blob: dd9bf7e410d2975f212accdf827df8a47c2d4c94 [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 */
20#include <linux/fs.h>
21#include <linux/file.h>
22#include <linux/pagemap.h>
Jens Axboed6b29d72007-06-04 09:59:47 +020023#include <linux/splice.h>
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -080024#include <linux/memcontrol.h>
Jens Axboe5274f052006-03-30 15:15:30 +020025#include <linux/mm_inline.h>
Jens Axboe5abc97a2006-03-30 15:16:46 +020026#include <linux/swap.h>
Jens Axboe4f6f0bd2006-04-02 23:04:46 +020027#include <linux/writeback.h>
Paul Gortmaker630d9c42011-11-16 23:57:37 -050028#include <linux/export.h>
Jens Axboe4f6f0bd2006-04-02 23:04:46 +020029#include <linux/syscalls.h>
Jens Axboe912d35f2006-04-26 10:59:21 +020030#include <linux/uio.h>
James Morris29ce2052007-07-13 11:44:32 +020031#include <linux/security.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090032#include <linux/gfp.h>
Eric Dumazet35f9c092012-04-05 03:05:35 +000033#include <linux/socket.h>
Al Viro76b021d2013-03-02 10:19:56 -050034#include <linux/compat.h>
Al Viro06ae43f2013-03-20 13:19:30 -040035#include "internal.h"
Jens Axboe5274f052006-03-30 15:15:30 +020036
Jens Axboe83f91352006-04-02 23:05:09 +020037/*
38 * Attempt to steal a page from a pipe buffer. This should perhaps go into
39 * a vm helper function, it's already simplified quite a bit by the
40 * addition of remove_mapping(). If success is returned, the caller may
41 * attempt to reuse this page for another destination.
42 */
Jens Axboe76ad4d12006-05-03 10:41:33 +020043static int page_cache_pipe_buf_steal(struct pipe_inode_info *pipe,
Jens Axboe5abc97a2006-03-30 15:16:46 +020044 struct pipe_buffer *buf)
45{
46 struct page *page = buf->page;
Jens Axboe9e94cd42006-06-20 15:01:12 +020047 struct address_space *mapping;
Jens Axboe5abc97a2006-03-30 15:16:46 +020048
Jens Axboe9e0267c2006-04-19 15:57:31 +020049 lock_page(page);
50
Jens Axboe9e94cd42006-06-20 15:01:12 +020051 mapping = page_mapping(page);
52 if (mapping) {
53 WARN_ON(!PageUptodate(page));
Jens Axboe5abc97a2006-03-30 15:16:46 +020054
Jens Axboe9e94cd42006-06-20 15:01:12 +020055 /*
56 * At least for ext2 with nobh option, we need to wait on
57 * writeback completing on this page, since we'll remove it
58 * from the pagecache. Otherwise truncate wont wait on the
59 * page, allowing the disk blocks to be reused by someone else
60 * before we actually wrote our data to them. fs corruption
61 * ensues.
62 */
63 wait_on_page_writeback(page);
Jens Axboead8d6f02006-04-02 23:10:32 +020064
David Howells266cf652009-04-03 16:42:36 +010065 if (page_has_private(page) &&
66 !try_to_release_page(page, GFP_KERNEL))
Jens Axboeca39d652008-05-20 21:27:41 +020067 goto out_unlock;
Jens Axboe4f6f0bd2006-04-02 23:04:46 +020068
Jens Axboe9e94cd42006-06-20 15:01:12 +020069 /*
70 * If we succeeded in removing the mapping, set LRU flag
71 * and return good.
72 */
73 if (remove_mapping(mapping, page)) {
74 buf->flags |= PIPE_BUF_FLAG_LRU;
75 return 0;
76 }
Jens Axboe9e0267c2006-04-19 15:57:31 +020077 }
Jens Axboe5abc97a2006-03-30 15:16:46 +020078
Jens Axboe9e94cd42006-06-20 15:01:12 +020079 /*
80 * Raced with truncate or failed to remove page from current
81 * address space, unlock and return failure.
82 */
Jens Axboeca39d652008-05-20 21:27:41 +020083out_unlock:
Jens Axboe9e94cd42006-06-20 15:01:12 +020084 unlock_page(page);
85 return 1;
Jens Axboe5abc97a2006-03-30 15:16:46 +020086}
87
Jens Axboe76ad4d12006-05-03 10:41:33 +020088static void page_cache_pipe_buf_release(struct pipe_inode_info *pipe,
Jens Axboe5274f052006-03-30 15:15:30 +020089 struct pipe_buffer *buf)
90{
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +030091 put_page(buf->page);
Jens Axboe14328732006-05-03 10:35:26 +020092 buf->flags &= ~PIPE_BUF_FLAG_LRU;
Jens Axboe5274f052006-03-30 15:15:30 +020093}
94
Jens Axboe08457182007-06-12 20:51:32 +020095/*
96 * Check whether the contents of buf is OK to access. Since the content
97 * is a page cache page, IO may be in flight.
98 */
Jens Axboecac36bb02007-06-14 13:10:48 +020099static int page_cache_pipe_buf_confirm(struct pipe_inode_info *pipe,
100 struct pipe_buffer *buf)
Jens Axboe5274f052006-03-30 15:15:30 +0200101{
102 struct page *page = buf->page;
Jens Axboe49d0b212006-04-10 09:04:41 +0200103 int err;
Jens Axboe5274f052006-03-30 15:15:30 +0200104
105 if (!PageUptodate(page)) {
Jens Axboe49d0b212006-04-10 09:04:41 +0200106 lock_page(page);
107
108 /*
109 * Page got truncated/unhashed. This will cause a 0-byte
Ingo Molnar73d62d82006-04-11 13:57:21 +0200110 * splice, if this is the first page.
Jens Axboe49d0b212006-04-10 09:04:41 +0200111 */
112 if (!page->mapping) {
113 err = -ENODATA;
114 goto error;
115 }
116
117 /*
Ingo Molnar73d62d82006-04-11 13:57:21 +0200118 * Uh oh, read-error from disk.
Jens Axboe49d0b212006-04-10 09:04:41 +0200119 */
120 if (!PageUptodate(page)) {
121 err = -EIO;
122 goto error;
123 }
124
125 /*
Jens Axboef84d7512006-05-01 19:59:03 +0200126 * Page is ok afterall, we are done.
Jens Axboe49d0b212006-04-10 09:04:41 +0200127 */
Jens Axboe5274f052006-03-30 15:15:30 +0200128 unlock_page(page);
Jens Axboe5274f052006-03-30 15:15:30 +0200129 }
130
Jens Axboef84d7512006-05-01 19:59:03 +0200131 return 0;
Jens Axboe49d0b212006-04-10 09:04:41 +0200132error:
133 unlock_page(page);
Jens Axboef84d7512006-05-01 19:59:03 +0200134 return err;
Jens Axboe70524492006-04-11 15:51:17 +0200135}
136
Hugh Dickins708e3502011-07-25 17:12:32 -0700137const struct pipe_buf_operations page_cache_pipe_buf_ops = {
Jens Axboe5274f052006-03-30 15:15:30 +0200138 .can_merge = 0,
Jens Axboecac36bb02007-06-14 13:10:48 +0200139 .confirm = page_cache_pipe_buf_confirm,
Jens Axboe5274f052006-03-30 15:15:30 +0200140 .release = page_cache_pipe_buf_release,
Jens Axboe5abc97a2006-03-30 15:16:46 +0200141 .steal = page_cache_pipe_buf_steal,
Jens Axboef84d7512006-05-01 19:59:03 +0200142 .get = generic_pipe_buf_get,
Jens Axboe5274f052006-03-30 15:15:30 +0200143};
144
Jens Axboe912d35f2006-04-26 10:59:21 +0200145static int user_page_pipe_buf_steal(struct pipe_inode_info *pipe,
146 struct pipe_buffer *buf)
147{
Jens Axboe7afa6fd2006-05-01 20:02:33 +0200148 if (!(buf->flags & PIPE_BUF_FLAG_GIFT))
149 return 1;
150
Jens Axboe14328732006-05-03 10:35:26 +0200151 buf->flags |= PIPE_BUF_FLAG_LRU;
Jens Axboe330ab712006-05-02 15:29:57 +0200152 return generic_pipe_buf_steal(pipe, buf);
Jens Axboe912d35f2006-04-26 10:59:21 +0200153}
154
Eric Dumazetd4c3cca2006-12-13 00:34:04 -0800155static const struct pipe_buf_operations user_page_pipe_buf_ops = {
Jens Axboe912d35f2006-04-26 10:59:21 +0200156 .can_merge = 0,
Jens Axboecac36bb02007-06-14 13:10:48 +0200157 .confirm = generic_pipe_buf_confirm,
Jens Axboe912d35f2006-04-26 10:59:21 +0200158 .release = page_cache_pipe_buf_release,
159 .steal = user_page_pipe_buf_steal,
Jens Axboef84d7512006-05-01 19:59:03 +0200160 .get = generic_pipe_buf_get,
Jens Axboe912d35f2006-04-26 10:59:21 +0200161};
162
Namhyung Kim825cdcb2011-05-23 19:58:53 +0200163static void wakeup_pipe_readers(struct pipe_inode_info *pipe)
164{
165 smp_mb();
166 if (waitqueue_active(&pipe->wait))
167 wake_up_interruptible(&pipe->wait);
168 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
169}
170
Jens Axboe932cc6d2007-06-21 13:10:21 +0200171/**
172 * splice_to_pipe - fill passed data into a pipe
173 * @pipe: pipe to fill
174 * @spd: data to fill
175 *
176 * Description:
Randy Dunlap79685b82007-07-27 08:08:51 +0200177 * @spd contains a map of pages and len/offset tuples, along with
Jens Axboe932cc6d2007-06-21 13:10:21 +0200178 * the struct pipe_buf_operations associated with these pages. This
179 * function will link that data to the pipe.
180 *
Jens Axboe83f91352006-04-02 23:05:09 +0200181 */
Jens Axboed6b29d72007-06-04 09:59:47 +0200182ssize_t splice_to_pipe(struct pipe_inode_info *pipe,
183 struct splice_pipe_desc *spd)
Jens Axboe5274f052006-03-30 15:15:30 +0200184{
Jens Axboe00de00b2007-06-15 13:14:22 +0200185 unsigned int spd_pages = spd->nr_pages;
Jens Axboe912d35f2006-04-26 10:59:21 +0200186 int ret, do_wakeup, page_nr;
Jens Axboe5274f052006-03-30 15:15:30 +0200187
Rabin Vincentd6785d92016-03-10 21:19:06 +0100188 if (!spd_pages)
189 return 0;
190
Jens Axboe5274f052006-03-30 15:15:30 +0200191 ret = 0;
192 do_wakeup = 0;
Jens Axboe912d35f2006-04-26 10:59:21 +0200193 page_nr = 0;
Jens Axboe5274f052006-03-30 15:15:30 +0200194
Miklos Szeredi61e0d472009-04-14 19:48:41 +0200195 pipe_lock(pipe);
Jens Axboe5274f052006-03-30 15:15:30 +0200196
Jens Axboe5274f052006-03-30 15:15:30 +0200197 for (;;) {
Ingo Molnar3a326a22006-04-10 15:18:35 +0200198 if (!pipe->readers) {
Jens Axboe5274f052006-03-30 15:15:30 +0200199 send_sig(SIGPIPE, current, 0);
200 if (!ret)
201 ret = -EPIPE;
202 break;
203 }
204
Jens Axboe35f3d142010-05-20 10:43:18 +0200205 if (pipe->nrbufs < pipe->buffers) {
206 int newbuf = (pipe->curbuf + pipe->nrbufs) & (pipe->buffers - 1);
Ingo Molnar3a326a22006-04-10 15:18:35 +0200207 struct pipe_buffer *buf = pipe->bufs + newbuf;
Jens Axboe5274f052006-03-30 15:15:30 +0200208
Jens Axboe912d35f2006-04-26 10:59:21 +0200209 buf->page = spd->pages[page_nr];
210 buf->offset = spd->partial[page_nr].offset;
211 buf->len = spd->partial[page_nr].len;
Jens Axboe497f9622007-06-11 12:00:45 +0200212 buf->private = spd->partial[page_nr].private;
Jens Axboe912d35f2006-04-26 10:59:21 +0200213 buf->ops = spd->ops;
Jens Axboe7afa6fd2006-05-01 20:02:33 +0200214 if (spd->flags & SPLICE_F_GIFT)
215 buf->flags |= PIPE_BUF_FLAG_GIFT;
216
Jens Axboe6f767b02006-04-11 13:53:56 +0200217 pipe->nrbufs++;
Jens Axboe912d35f2006-04-26 10:59:21 +0200218 page_nr++;
219 ret += buf->len;
220
Al Viro6447a3c2013-03-21 11:01:38 -0400221 if (pipe->files)
Jens Axboe6f767b02006-04-11 13:53:56 +0200222 do_wakeup = 1;
Jens Axboe5274f052006-03-30 15:15:30 +0200223
Jens Axboe912d35f2006-04-26 10:59:21 +0200224 if (!--spd->nr_pages)
Jens Axboe5274f052006-03-30 15:15:30 +0200225 break;
Jens Axboe35f3d142010-05-20 10:43:18 +0200226 if (pipe->nrbufs < pipe->buffers)
Jens Axboe5274f052006-03-30 15:15:30 +0200227 continue;
228
229 break;
230 }
231
Jens Axboe912d35f2006-04-26 10:59:21 +0200232 if (spd->flags & SPLICE_F_NONBLOCK) {
Linus Torvalds29e35092006-04-02 12:46:35 -0700233 if (!ret)
234 ret = -EAGAIN;
235 break;
236 }
237
Jens Axboe5274f052006-03-30 15:15:30 +0200238 if (signal_pending(current)) {
239 if (!ret)
240 ret = -ERESTARTSYS;
241 break;
242 }
243
244 if (do_wakeup) {
Jens Axboec0bd1f62006-04-10 09:03:32 +0200245 smp_mb();
Ingo Molnar3a326a22006-04-10 15:18:35 +0200246 if (waitqueue_active(&pipe->wait))
247 wake_up_interruptible_sync(&pipe->wait);
248 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
Jens Axboe5274f052006-03-30 15:15:30 +0200249 do_wakeup = 0;
250 }
251
Ingo Molnar3a326a22006-04-10 15:18:35 +0200252 pipe->waiting_writers++;
253 pipe_wait(pipe);
254 pipe->waiting_writers--;
Jens Axboe5274f052006-03-30 15:15:30 +0200255 }
256
Miklos Szeredi61e0d472009-04-14 19:48:41 +0200257 pipe_unlock(pipe);
Jens Axboe5274f052006-03-30 15:15:30 +0200258
Namhyung Kim825cdcb2011-05-23 19:58:53 +0200259 if (do_wakeup)
260 wakeup_pipe_readers(pipe);
Jens Axboe5274f052006-03-30 15:15:30 +0200261
Jens Axboe00de00b2007-06-15 13:14:22 +0200262 while (page_nr < spd_pages)
Jens Axboebbdfc2f2007-11-06 23:29:47 -0800263 spd->spd_release(spd, page_nr++);
Jens Axboe5274f052006-03-30 15:15:30 +0200264
265 return ret;
266}
Hannes Frederic Sowa2b514572015-05-21 17:00:01 +0200267EXPORT_SYMBOL_GPL(splice_to_pipe);
Jens Axboe5274f052006-03-30 15:15:30 +0200268
Hugh Dickins708e3502011-07-25 17:12:32 -0700269void spd_release_page(struct splice_pipe_desc *spd, unsigned int i)
Jens Axboebbdfc2f2007-11-06 23:29:47 -0800270{
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300271 put_page(spd->pages[i]);
Jens Axboebbdfc2f2007-11-06 23:29:47 -0800272}
273
Jens Axboe35f3d142010-05-20 10:43:18 +0200274/*
275 * Check if we need to grow the arrays holding pages and partial page
276 * descriptions.
277 */
Eric Dumazet047fe362012-06-12 15:24:40 +0200278int splice_grow_spd(const struct pipe_inode_info *pipe, struct splice_pipe_desc *spd)
Jens Axboe35f3d142010-05-20 10:43:18 +0200279{
Eric Dumazet047fe362012-06-12 15:24:40 +0200280 unsigned int buffers = ACCESS_ONCE(pipe->buffers);
281
282 spd->nr_pages_max = buffers;
283 if (buffers <= PIPE_DEF_BUFFERS)
Jens Axboe35f3d142010-05-20 10:43:18 +0200284 return 0;
285
Eric Dumazet047fe362012-06-12 15:24:40 +0200286 spd->pages = kmalloc(buffers * sizeof(struct page *), GFP_KERNEL);
287 spd->partial = kmalloc(buffers * sizeof(struct partial_page), GFP_KERNEL);
Jens Axboe35f3d142010-05-20 10:43:18 +0200288
289 if (spd->pages && spd->partial)
290 return 0;
291
292 kfree(spd->pages);
293 kfree(spd->partial);
294 return -ENOMEM;
295}
296
Eric Dumazet047fe362012-06-12 15:24:40 +0200297void splice_shrink_spd(struct splice_pipe_desc *spd)
Jens Axboe35f3d142010-05-20 10:43:18 +0200298{
Eric Dumazet047fe362012-06-12 15:24:40 +0200299 if (spd->nr_pages_max <= PIPE_DEF_BUFFERS)
Jens Axboe35f3d142010-05-20 10:43:18 +0200300 return;
301
302 kfree(spd->pages);
303 kfree(spd->partial);
304}
305
Ingo Molnar3a326a22006-04-10 15:18:35 +0200306static int
Jens Axboecbb7e572006-04-11 14:57:50 +0200307__generic_file_splice_read(struct file *in, loff_t *ppos,
308 struct pipe_inode_info *pipe, size_t len,
309 unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +0200310{
311 struct address_space *mapping = in->f_mapping;
Fengguang Wud8983912007-07-19 01:48:06 -0700312 unsigned int loff, nr_pages, req_pages;
Jens Axboe35f3d142010-05-20 10:43:18 +0200313 struct page *pages[PIPE_DEF_BUFFERS];
314 struct partial_page partial[PIPE_DEF_BUFFERS];
Jens Axboe5274f052006-03-30 15:15:30 +0200315 struct page *page;
Jens Axboe91ad66e2006-04-19 15:55:10 +0200316 pgoff_t index, end_index;
317 loff_t isize;
Jens Axboeeb207962006-04-27 11:05:22 +0200318 int error, page_nr;
Jens Axboe912d35f2006-04-26 10:59:21 +0200319 struct splice_pipe_desc spd = {
320 .pages = pages,
321 .partial = partial,
Eric Dumazet047fe362012-06-12 15:24:40 +0200322 .nr_pages_max = PIPE_DEF_BUFFERS,
Jens Axboe912d35f2006-04-26 10:59:21 +0200323 .flags = flags,
324 .ops = &page_cache_pipe_buf_ops,
Jens Axboebbdfc2f2007-11-06 23:29:47 -0800325 .spd_release = spd_release_page,
Jens Axboe912d35f2006-04-26 10:59:21 +0200326 };
Jens Axboe5274f052006-03-30 15:15:30 +0200327
Jens Axboe35f3d142010-05-20 10:43:18 +0200328 if (splice_grow_spd(pipe, &spd))
329 return -ENOMEM;
330
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300331 index = *ppos >> PAGE_SHIFT;
332 loff = *ppos & ~PAGE_MASK;
333 req_pages = (len + loff + PAGE_SIZE - 1) >> PAGE_SHIFT;
Eric Dumazet047fe362012-06-12 15:24:40 +0200334 nr_pages = min(req_pages, spd.nr_pages_max);
Jens Axboe5274f052006-03-30 15:15:30 +0200335
336 /*
Jens Axboeeb207962006-04-27 11:05:22 +0200337 * Lookup the (hopefully) full range of pages we need.
338 */
Jens Axboe35f3d142010-05-20 10:43:18 +0200339 spd.nr_pages = find_get_pages_contig(mapping, index, nr_pages, spd.pages);
Fengguang Wu431a48202007-07-19 01:48:05 -0700340 index += spd.nr_pages;
Jens Axboeeb207962006-04-27 11:05:22 +0200341
342 /*
343 * If find_get_pages_contig() returned fewer pages than we needed,
Fengguang Wu431a48202007-07-19 01:48:05 -0700344 * readahead/allocate the rest and fill in the holes.
Jens Axboeeb207962006-04-27 11:05:22 +0200345 */
Fengguang Wu431a48202007-07-19 01:48:05 -0700346 if (spd.nr_pages < nr_pages)
Rusty Russellcf914a72007-07-19 01:48:08 -0700347 page_cache_sync_readahead(mapping, &in->f_ra, in,
348 index, req_pages - spd.nr_pages);
Fengguang Wu431a48202007-07-19 01:48:05 -0700349
Jens Axboe932cc6d2007-06-21 13:10:21 +0200350 error = 0;
Jens Axboeeb207962006-04-27 11:05:22 +0200351 while (spd.nr_pages < nr_pages) {
352 /*
353 * Page could be there, find_get_pages_contig() breaks on
354 * the first hole.
355 */
356 page = find_get_page(mapping, index);
357 if (!page) {
Jens Axboee27dedd2006-05-01 19:59:54 +0200358 /*
Jens Axboeeb207962006-04-27 11:05:22 +0200359 * page didn't exist, allocate one.
360 */
361 page = page_cache_alloc_cold(mapping);
362 if (!page)
363 break;
364
365 error = add_to_page_cache_lru(page, mapping, index,
Michal Hockoc62d2552015-11-06 16:28:49 -0800366 mapping_gfp_constraint(mapping, GFP_KERNEL));
Jens Axboeeb207962006-04-27 11:05:22 +0200367 if (unlikely(error)) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300368 put_page(page);
Jens Axboea0548872006-05-03 10:58:22 +0200369 if (error == -EEXIST)
370 continue;
Jens Axboeeb207962006-04-27 11:05:22 +0200371 break;
372 }
373 /*
374 * add_to_page_cache() locks the page, unlock it
375 * to avoid convoluting the logic below even more.
376 */
377 unlock_page(page);
378 }
379
Jens Axboe35f3d142010-05-20 10:43:18 +0200380 spd.pages[spd.nr_pages++] = page;
Jens Axboeeb207962006-04-27 11:05:22 +0200381 index++;
382 }
383
384 /*
385 * Now loop over the map and see if we need to start IO on any
386 * pages, fill in the partial map, etc.
387 */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300388 index = *ppos >> PAGE_SHIFT;
Jens Axboeeb207962006-04-27 11:05:22 +0200389 nr_pages = spd.nr_pages;
390 spd.nr_pages = 0;
391 for (page_nr = 0; page_nr < nr_pages; page_nr++) {
Jens Axboe82aa5d62006-04-20 13:05:48 +0200392 unsigned int this_len;
393
394 if (!len)
395 break;
396
397 /*
398 * this_len is the max we'll use from this page
399 */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300400 this_len = min_t(unsigned long, len, PAGE_SIZE - loff);
Jens Axboe35f3d142010-05-20 10:43:18 +0200401 page = spd.pages[page_nr];
Jens Axboe7480a902006-04-11 13:52:47 +0200402
Fengguang Wua08a1662007-07-19 01:48:03 -0700403 if (PageReadahead(page))
Rusty Russellcf914a72007-07-19 01:48:08 -0700404 page_cache_async_readahead(mapping, &in->f_ra, in,
Fengguang Wud8983912007-07-19 01:48:06 -0700405 page, index, req_pages - page_nr);
Fengguang Wua08a1662007-07-19 01:48:03 -0700406
Jens Axboe7480a902006-04-11 13:52:47 +0200407 /*
408 * If the page isn't uptodate, we may need to start io on it
409 */
410 if (!PageUptodate(page)) {
Miklos Szeredi69650312010-08-03 12:48:50 +0200411 lock_page(page);
Jens Axboe7480a902006-04-11 13:52:47 +0200412
413 /*
Miklos Szeredi32502b82008-07-04 09:35:17 +0200414 * Page was truncated, or invalidated by the
415 * filesystem. Redo the find/create, but this time the
416 * page is kept locked, so there's no chance of another
417 * race with truncate/invalidate.
Jens Axboe7480a902006-04-11 13:52:47 +0200418 */
419 if (!page->mapping) {
420 unlock_page(page);
Abhi Das90330e62015-12-18 14:11:36 -0600421retry_lookup:
Miklos Szeredi32502b82008-07-04 09:35:17 +0200422 page = find_or_create_page(mapping, index,
423 mapping_gfp_mask(mapping));
424
425 if (!page) {
426 error = -ENOMEM;
427 break;
428 }
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300429 put_page(spd.pages[page_nr]);
Jens Axboe35f3d142010-05-20 10:43:18 +0200430 spd.pages[page_nr] = page;
Jens Axboe7480a902006-04-11 13:52:47 +0200431 }
432 /*
433 * page was already under io and is now done, great
434 */
435 if (PageUptodate(page)) {
436 unlock_page(page);
437 goto fill_it;
438 }
439
Jens Axboe7480a902006-04-11 13:52:47 +0200440 /*
441 * need to read in the page
442 */
443 error = mapping->a_ops->readpage(in, page);
Jens Axboe7480a902006-04-11 13:52:47 +0200444 if (unlikely(error)) {
Jens Axboeeb207962006-04-27 11:05:22 +0200445 /*
Abhi Das90330e62015-12-18 14:11:36 -0600446 * Re-lookup the page
Jens Axboeeb207962006-04-27 11:05:22 +0200447 */
Jens Axboe7480a902006-04-11 13:52:47 +0200448 if (error == AOP_TRUNCATED_PAGE)
Abhi Das90330e62015-12-18 14:11:36 -0600449 goto retry_lookup;
Jens Axboeeb207962006-04-27 11:05:22 +0200450
Jens Axboe7480a902006-04-11 13:52:47 +0200451 break;
452 }
Jens Axboe620a3242007-06-07 09:39:42 +0200453 }
454fill_it:
455 /*
456 * i_size must be checked after PageUptodate.
457 */
458 isize = i_size_read(mapping->host);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300459 end_index = (isize - 1) >> PAGE_SHIFT;
Jens Axboe620a3242007-06-07 09:39:42 +0200460 if (unlikely(!isize || index > end_index))
461 break;
462
463 /*
464 * if this is the last page, see if we need to shrink
465 * the length and stop
466 */
467 if (end_index == index) {
468 unsigned int plen;
Jens Axboe91ad66e2006-04-19 15:55:10 +0200469
470 /*
Jens Axboe620a3242007-06-07 09:39:42 +0200471 * max good bytes in this page
Jens Axboe91ad66e2006-04-19 15:55:10 +0200472 */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300473 plen = ((isize - 1) & ~PAGE_MASK) + 1;
Jens Axboe620a3242007-06-07 09:39:42 +0200474 if (plen <= loff)
Jens Axboe91ad66e2006-04-19 15:55:10 +0200475 break;
Jens Axboe91ad66e2006-04-19 15:55:10 +0200476
477 /*
Jens Axboe620a3242007-06-07 09:39:42 +0200478 * force quit after adding this page
Jens Axboe91ad66e2006-04-19 15:55:10 +0200479 */
Jens Axboe620a3242007-06-07 09:39:42 +0200480 this_len = min(this_len, plen - loff);
481 len = this_len;
Jens Axboe7480a902006-04-11 13:52:47 +0200482 }
Jens Axboe620a3242007-06-07 09:39:42 +0200483
Jens Axboe35f3d142010-05-20 10:43:18 +0200484 spd.partial[page_nr].offset = loff;
485 spd.partial[page_nr].len = this_len;
Jens Axboe82aa5d62006-04-20 13:05:48 +0200486 len -= this_len;
Jens Axboe91ad66e2006-04-19 15:55:10 +0200487 loff = 0;
Jens Axboeeb207962006-04-27 11:05:22 +0200488 spd.nr_pages++;
489 index++;
Jens Axboe5274f052006-03-30 15:15:30 +0200490 }
491
Jens Axboeeb207962006-04-27 11:05:22 +0200492 /*
Hugh Dickins475ecad2007-06-07 09:36:00 +0200493 * Release any pages at the end, if we quit early. 'page_nr' is how far
Jens Axboeeb207962006-04-27 11:05:22 +0200494 * we got, 'nr_pages' is how many pages are in the map.
495 */
496 while (page_nr < nr_pages)
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300497 put_page(spd.pages[page_nr++]);
498 in->f_ra.prev_pos = (loff_t)index << PAGE_SHIFT;
Jens Axboeeb207962006-04-27 11:05:22 +0200499
Jens Axboe912d35f2006-04-26 10:59:21 +0200500 if (spd.nr_pages)
Jens Axboe35f3d142010-05-20 10:43:18 +0200501 error = splice_to_pipe(pipe, &spd);
Jens Axboe5274f052006-03-30 15:15:30 +0200502
Eric Dumazet047fe362012-06-12 15:24:40 +0200503 splice_shrink_spd(&spd);
Jens Axboe7480a902006-04-11 13:52:47 +0200504 return error;
Jens Axboe5274f052006-03-30 15:15:30 +0200505}
506
Jens Axboe83f91352006-04-02 23:05:09 +0200507/**
508 * generic_file_splice_read - splice data from file to a pipe
509 * @in: file to splice from
Jens Axboe932cc6d2007-06-21 13:10:21 +0200510 * @ppos: position in @in
Jens Axboe83f91352006-04-02 23:05:09 +0200511 * @pipe: pipe to splice to
512 * @len: number of bytes to splice
513 * @flags: splice modifier flags
514 *
Jens Axboe932cc6d2007-06-21 13:10:21 +0200515 * Description:
516 * Will read pages from given file and fill them into a pipe. Can be
517 * used as long as the address_space operations for the source implements
518 * a readpage() hook.
519 *
Jens Axboe83f91352006-04-02 23:05:09 +0200520 */
Jens Axboecbb7e572006-04-11 14:57:50 +0200521ssize_t generic_file_splice_read(struct file *in, loff_t *ppos,
522 struct pipe_inode_info *pipe, size_t len,
523 unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +0200524{
Jens Axboed366d3982007-06-01 14:54:11 +0200525 loff_t isize, left;
Jens Axboe8191ecd2008-04-10 08:24:25 +0200526 int ret;
Jens Axboed366d3982007-06-01 14:54:11 +0200527
Boaz Harroshbe64f882015-04-15 16:15:17 -0700528 if (IS_DAX(in->f_mapping->host))
529 return default_file_splice_read(in, ppos, pipe, len, flags);
530
Jens Axboed366d3982007-06-01 14:54:11 +0200531 isize = i_size_read(in->f_mapping->host);
532 if (unlikely(*ppos >= isize))
533 return 0;
534
535 left = isize - *ppos;
536 if (unlikely(left < len))
537 len = left;
Jens Axboe5274f052006-03-30 15:15:30 +0200538
Jens Axboe8191ecd2008-04-10 08:24:25 +0200539 ret = __generic_file_splice_read(in, ppos, pipe, len, flags);
Miklos Szeredi723590e2009-08-15 08:43:22 +0200540 if (ret > 0) {
Jens Axboecbb7e572006-04-11 14:57:50 +0200541 *ppos += ret;
Miklos Szeredi723590e2009-08-15 08:43:22 +0200542 file_accessed(in);
543 }
Jens Axboe5274f052006-03-30 15:15:30 +0200544
545 return ret;
546}
Jens Axboe059a8f32006-04-02 23:06:05 +0200547EXPORT_SYMBOL(generic_file_splice_read);
548
Miklos Szeredi68181732009-05-07 15:37:36 +0200549static const struct pipe_buf_operations default_pipe_buf_ops = {
550 .can_merge = 0,
Miklos Szeredi68181732009-05-07 15:37:36 +0200551 .confirm = generic_pipe_buf_confirm,
552 .release = generic_pipe_buf_release,
553 .steal = generic_pipe_buf_steal,
554 .get = generic_pipe_buf_get,
555};
556
Miklos Szeredi28a625c2014-01-22 19:36:57 +0100557static int generic_pipe_buf_nosteal(struct pipe_inode_info *pipe,
558 struct pipe_buffer *buf)
559{
560 return 1;
561}
562
563/* Pipe buffer operations for a socket and similar. */
564const struct pipe_buf_operations nosteal_pipe_buf_ops = {
565 .can_merge = 0,
Miklos Szeredi28a625c2014-01-22 19:36:57 +0100566 .confirm = generic_pipe_buf_confirm,
567 .release = generic_pipe_buf_release,
568 .steal = generic_pipe_buf_nosteal,
569 .get = generic_pipe_buf_get,
570};
571EXPORT_SYMBOL(nosteal_pipe_buf_ops);
572
Miklos Szeredi68181732009-05-07 15:37:36 +0200573static ssize_t kernel_readv(struct file *file, const struct iovec *vec,
574 unsigned long vlen, loff_t offset)
575{
576 mm_segment_t old_fs;
577 loff_t pos = offset;
578 ssize_t res;
579
580 old_fs = get_fs();
581 set_fs(get_ds());
582 /* The cast to a user pointer is valid due to the set_fs() */
Christoph Hellwig793b80e2016-03-03 16:03:58 +0100583 res = vfs_readv(file, (const struct iovec __user *)vec, vlen, &pos, 0);
Miklos Szeredi68181732009-05-07 15:37:36 +0200584 set_fs(old_fs);
585
586 return res;
587}
588
Al Viro7bb307e2013-02-23 14:51:48 -0500589ssize_t kernel_write(struct file *file, const char *buf, size_t count,
Miklos Szeredib2858d72009-05-19 11:37:46 +0200590 loff_t pos)
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +0200591{
592 mm_segment_t old_fs;
593 ssize_t res;
594
595 old_fs = get_fs();
596 set_fs(get_ds());
597 /* The cast to a user pointer is valid due to the set_fs() */
Al Viro7bb307e2013-02-23 14:51:48 -0500598 res = vfs_write(file, (__force const char __user *)buf, count, &pos);
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +0200599 set_fs(old_fs);
600
601 return res;
602}
Al Viro7bb307e2013-02-23 14:51:48 -0500603EXPORT_SYMBOL(kernel_write);
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +0200604
Miklos Szeredi68181732009-05-07 15:37:36 +0200605ssize_t default_file_splice_read(struct file *in, loff_t *ppos,
606 struct pipe_inode_info *pipe, size_t len,
607 unsigned int flags)
608{
609 unsigned int nr_pages;
610 unsigned int nr_freed;
611 size_t offset;
Jens Axboe35f3d142010-05-20 10:43:18 +0200612 struct page *pages[PIPE_DEF_BUFFERS];
613 struct partial_page partial[PIPE_DEF_BUFFERS];
614 struct iovec *vec, __vec[PIPE_DEF_BUFFERS];
Miklos Szeredi68181732009-05-07 15:37:36 +0200615 ssize_t res;
616 size_t this_len;
617 int error;
618 int i;
619 struct splice_pipe_desc spd = {
620 .pages = pages,
621 .partial = partial,
Eric Dumazet047fe362012-06-12 15:24:40 +0200622 .nr_pages_max = PIPE_DEF_BUFFERS,
Miklos Szeredi68181732009-05-07 15:37:36 +0200623 .flags = flags,
624 .ops = &default_pipe_buf_ops,
625 .spd_release = spd_release_page,
626 };
627
Jens Axboe35f3d142010-05-20 10:43:18 +0200628 if (splice_grow_spd(pipe, &spd))
629 return -ENOMEM;
630
631 res = -ENOMEM;
632 vec = __vec;
Eric Dumazet047fe362012-06-12 15:24:40 +0200633 if (spd.nr_pages_max > PIPE_DEF_BUFFERS) {
634 vec = kmalloc(spd.nr_pages_max * sizeof(struct iovec), GFP_KERNEL);
Jens Axboe35f3d142010-05-20 10:43:18 +0200635 if (!vec)
636 goto shrink_ret;
637 }
638
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300639 offset = *ppos & ~PAGE_MASK;
640 nr_pages = (len + offset + PAGE_SIZE - 1) >> PAGE_SHIFT;
Miklos Szeredi68181732009-05-07 15:37:36 +0200641
Eric Dumazet047fe362012-06-12 15:24:40 +0200642 for (i = 0; i < nr_pages && i < spd.nr_pages_max && len; i++) {
Miklos Szeredi68181732009-05-07 15:37:36 +0200643 struct page *page;
644
Jens Axboe4f231222009-05-13 08:35:35 +0200645 page = alloc_page(GFP_USER);
Miklos Szeredi68181732009-05-07 15:37:36 +0200646 error = -ENOMEM;
647 if (!page)
648 goto err;
649
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300650 this_len = min_t(size_t, len, PAGE_SIZE - offset);
Jens Axboe4f231222009-05-13 08:35:35 +0200651 vec[i].iov_base = (void __user *) page_address(page);
Miklos Szeredi68181732009-05-07 15:37:36 +0200652 vec[i].iov_len = this_len;
Jens Axboe35f3d142010-05-20 10:43:18 +0200653 spd.pages[i] = page;
Miklos Szeredi68181732009-05-07 15:37:36 +0200654 spd.nr_pages++;
655 len -= this_len;
656 offset = 0;
657 }
658
659 res = kernel_readv(in, vec, spd.nr_pages, *ppos);
Andrew Morton77f6bf52009-05-14 09:49:44 +0200660 if (res < 0) {
661 error = res;
Miklos Szeredi68181732009-05-07 15:37:36 +0200662 goto err;
Andrew Morton77f6bf52009-05-14 09:49:44 +0200663 }
Miklos Szeredi68181732009-05-07 15:37:36 +0200664
665 error = 0;
666 if (!res)
667 goto err;
668
669 nr_freed = 0;
670 for (i = 0; i < spd.nr_pages; i++) {
Miklos Szeredi68181732009-05-07 15:37:36 +0200671 this_len = min_t(size_t, vec[i].iov_len, res);
Jens Axboe35f3d142010-05-20 10:43:18 +0200672 spd.partial[i].offset = 0;
673 spd.partial[i].len = this_len;
Miklos Szeredi68181732009-05-07 15:37:36 +0200674 if (!this_len) {
Jens Axboe35f3d142010-05-20 10:43:18 +0200675 __free_page(spd.pages[i]);
676 spd.pages[i] = NULL;
Miklos Szeredi68181732009-05-07 15:37:36 +0200677 nr_freed++;
678 }
679 res -= this_len;
680 }
681 spd.nr_pages -= nr_freed;
682
683 res = splice_to_pipe(pipe, &spd);
684 if (res > 0)
685 *ppos += res;
686
Jens Axboe35f3d142010-05-20 10:43:18 +0200687shrink_ret:
688 if (vec != __vec)
689 kfree(vec);
Eric Dumazet047fe362012-06-12 15:24:40 +0200690 splice_shrink_spd(&spd);
Miklos Szeredi68181732009-05-07 15:37:36 +0200691 return res;
692
693err:
Jens Axboe4f231222009-05-13 08:35:35 +0200694 for (i = 0; i < spd.nr_pages; i++)
Jens Axboe35f3d142010-05-20 10:43:18 +0200695 __free_page(spd.pages[i]);
Jens Axboe4f231222009-05-13 08:35:35 +0200696
Jens Axboe35f3d142010-05-20 10:43:18 +0200697 res = error;
698 goto shrink_ret;
Miklos Szeredi68181732009-05-07 15:37:36 +0200699}
700EXPORT_SYMBOL(default_file_splice_read);
701
Jens Axboe5274f052006-03-30 15:15:30 +0200702/*
Jens Axboe4f6f0bd2006-04-02 23:04:46 +0200703 * Send 'sd->len' bytes to socket from 'sd->file' at position 'sd->pos'
Jens Axboe016b6612006-04-25 15:42:00 +0200704 * using sendpage(). Return the number of bytes sent.
Jens Axboe5274f052006-03-30 15:15:30 +0200705 */
Jens Axboe76ad4d12006-05-03 10:41:33 +0200706static int pipe_to_sendpage(struct pipe_inode_info *pipe,
Jens Axboe5274f052006-03-30 15:15:30 +0200707 struct pipe_buffer *buf, struct splice_desc *sd)
708{
Jens Axboe6a14b902007-06-14 13:08:55 +0200709 struct file *file = sd->u.file;
Jens Axboe5274f052006-03-30 15:15:30 +0200710 loff_t pos = sd->pos;
Michał Mirosława8adbe32010-12-17 08:56:44 +0100711 int more;
Jens Axboe5274f052006-03-30 15:15:30 +0200712
Al Viro72c2d532013-09-22 16:27:52 -0400713 if (!likely(file->f_op->sendpage))
Michał Mirosława8adbe32010-12-17 08:56:44 +0100714 return -EINVAL;
Jens Axboe5274f052006-03-30 15:15:30 +0200715
Eric Dumazet35f9c092012-04-05 03:05:35 +0000716 more = (sd->flags & SPLICE_F_MORE) ? MSG_MORE : 0;
Eric Dumazetae62ca72013-01-06 18:21:49 +0000717
718 if (sd->len < sd->total_len && pipe->nrbufs > 1)
Eric Dumazet35f9c092012-04-05 03:05:35 +0000719 more |= MSG_SENDPAGE_NOTLAST;
Eric Dumazetae62ca72013-01-06 18:21:49 +0000720
Michał Mirosława8adbe32010-12-17 08:56:44 +0100721 return file->f_op->sendpage(file, buf->page, buf->offset,
722 sd->len, &pos, more);
Jens Axboe5274f052006-03-30 15:15:30 +0200723}
724
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200725static void wakeup_pipe_writers(struct pipe_inode_info *pipe)
726{
727 smp_mb();
728 if (waitqueue_active(&pipe->wait))
729 wake_up_interruptible(&pipe->wait);
730 kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
731}
732
733/**
734 * splice_from_pipe_feed - feed available data from a pipe to a file
735 * @pipe: pipe to splice from
736 * @sd: information to @actor
737 * @actor: handler that splices the data
738 *
739 * Description:
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200740 * This function loops over the pipe and calls @actor to do the
741 * actual moving of a single struct pipe_buffer to the desired
742 * destination. It returns when there's no more buffers left in
743 * the pipe or if the requested number of bytes (@sd->total_len)
744 * have been copied. It returns a positive number (one) if the
745 * pipe needs to be filled with more data, zero if the required
746 * number of bytes have been copied and -errno on error.
747 *
748 * This, together with splice_from_pipe_{begin,end,next}, may be
749 * used to implement the functionality of __splice_from_pipe() when
750 * locking is required around copying the pipe buffers to the
751 * destination.
752 */
Al Viro96f9bc82014-04-05 04:35:49 -0400753static int splice_from_pipe_feed(struct pipe_inode_info *pipe, struct splice_desc *sd,
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200754 splice_actor *actor)
755{
756 int ret;
757
758 while (pipe->nrbufs) {
759 struct pipe_buffer *buf = pipe->bufs + pipe->curbuf;
760 const struct pipe_buf_operations *ops = buf->ops;
761
762 sd->len = buf->len;
763 if (sd->len > sd->total_len)
764 sd->len = sd->total_len;
765
Michał Mirosława8adbe32010-12-17 08:56:44 +0100766 ret = buf->ops->confirm(pipe, buf);
767 if (unlikely(ret)) {
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200768 if (ret == -ENODATA)
769 ret = 0;
770 return ret;
771 }
Michał Mirosława8adbe32010-12-17 08:56:44 +0100772
773 ret = actor(pipe, buf, sd);
774 if (ret <= 0)
775 return ret;
776
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200777 buf->offset += ret;
778 buf->len -= ret;
779
780 sd->num_spliced += ret;
781 sd->len -= ret;
782 sd->pos += ret;
783 sd->total_len -= ret;
784
785 if (!buf->len) {
786 buf->ops = NULL;
787 ops->release(pipe, buf);
Jens Axboe35f3d142010-05-20 10:43:18 +0200788 pipe->curbuf = (pipe->curbuf + 1) & (pipe->buffers - 1);
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200789 pipe->nrbufs--;
Al Viro6447a3c2013-03-21 11:01:38 -0400790 if (pipe->files)
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200791 sd->need_wakeup = true;
792 }
793
794 if (!sd->total_len)
795 return 0;
796 }
797
798 return 1;
799}
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200800
801/**
802 * splice_from_pipe_next - wait for some data to splice from
803 * @pipe: pipe to splice from
804 * @sd: information about the splice operation
805 *
806 * Description:
807 * This function will wait for some data and return a positive
808 * value (one) if pipe buffers are available. It will return zero
809 * or -errno if no more data needs to be spliced.
810 */
Al Viro96f9bc82014-04-05 04:35:49 -0400811static int splice_from_pipe_next(struct pipe_inode_info *pipe, struct splice_desc *sd)
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200812{
Jan Karac725bfc2015-11-23 13:09:50 +0100813 /*
814 * Check for signal early to make process killable when there are
815 * always buffers available
816 */
817 if (signal_pending(current))
818 return -ERESTARTSYS;
819
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200820 while (!pipe->nrbufs) {
821 if (!pipe->writers)
822 return 0;
823
824 if (!pipe->waiting_writers && sd->num_spliced)
825 return 0;
826
827 if (sd->flags & SPLICE_F_NONBLOCK)
828 return -EAGAIN;
829
830 if (signal_pending(current))
831 return -ERESTARTSYS;
832
833 if (sd->need_wakeup) {
834 wakeup_pipe_writers(pipe);
835 sd->need_wakeup = false;
836 }
837
838 pipe_wait(pipe);
839 }
840
841 return 1;
842}
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200843
844/**
845 * splice_from_pipe_begin - start splicing from pipe
Randy Dunlapb80901b2009-04-16 19:09:55 -0700846 * @sd: information about the splice operation
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200847 *
848 * Description:
849 * This function should be called before a loop containing
850 * splice_from_pipe_next() and splice_from_pipe_feed() to
851 * initialize the necessary fields of @sd.
852 */
Al Viro96f9bc82014-04-05 04:35:49 -0400853static void splice_from_pipe_begin(struct splice_desc *sd)
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200854{
855 sd->num_spliced = 0;
856 sd->need_wakeup = false;
857}
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200858
859/**
860 * splice_from_pipe_end - finish splicing from pipe
861 * @pipe: pipe to splice from
862 * @sd: information about the splice operation
863 *
864 * Description:
865 * This function will wake up pipe writers if necessary. It should
866 * be called after a loop containing splice_from_pipe_next() and
867 * splice_from_pipe_feed().
868 */
Al Viro96f9bc82014-04-05 04:35:49 -0400869static void splice_from_pipe_end(struct pipe_inode_info *pipe, struct splice_desc *sd)
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200870{
871 if (sd->need_wakeup)
872 wakeup_pipe_writers(pipe);
873}
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200874
Jens Axboe932cc6d2007-06-21 13:10:21 +0200875/**
876 * __splice_from_pipe - splice data from a pipe to given actor
877 * @pipe: pipe to splice from
878 * @sd: information to @actor
879 * @actor: handler that splices the data
880 *
881 * Description:
882 * This function does little more than loop over the pipe and call
883 * @actor to do the actual moving of a single struct pipe_buffer to
884 * the desired destination. See pipe_to_file, pipe_to_sendpage, or
885 * pipe_to_user.
886 *
Jens Axboe83f91352006-04-02 23:05:09 +0200887 */
Jens Axboec66ab6f2007-06-12 21:17:17 +0200888ssize_t __splice_from_pipe(struct pipe_inode_info *pipe, struct splice_desc *sd,
889 splice_actor *actor)
Jens Axboe5274f052006-03-30 15:15:30 +0200890{
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200891 int ret;
Jens Axboe5274f052006-03-30 15:15:30 +0200892
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200893 splice_from_pipe_begin(sd);
894 do {
Jan Karac2489e02015-11-23 13:09:51 +0100895 cond_resched();
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200896 ret = splice_from_pipe_next(pipe, sd);
897 if (ret > 0)
898 ret = splice_from_pipe_feed(pipe, sd, actor);
899 } while (ret > 0);
900 splice_from_pipe_end(pipe, sd);
Jens Axboe5274f052006-03-30 15:15:30 +0200901
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200902 return sd->num_spliced ? sd->num_spliced : ret;
Jens Axboe5274f052006-03-30 15:15:30 +0200903}
Mark Fasheh40bee44e2007-03-21 13:11:02 +0100904EXPORT_SYMBOL(__splice_from_pipe);
Jens Axboe5274f052006-03-30 15:15:30 +0200905
Jens Axboe932cc6d2007-06-21 13:10:21 +0200906/**
907 * splice_from_pipe - splice data from a pipe to a file
908 * @pipe: pipe to splice from
909 * @out: file to splice to
910 * @ppos: position in @out
911 * @len: how many bytes to splice
912 * @flags: splice modifier flags
913 * @actor: handler that splices the data
914 *
915 * Description:
Miklos Szeredi29339702009-04-14 19:48:37 +0200916 * See __splice_from_pipe. This function locks the pipe inode,
Jens Axboe932cc6d2007-06-21 13:10:21 +0200917 * otherwise it's identical to __splice_from_pipe().
918 *
919 */
Mark Fasheh6da61802006-10-17 18:43:07 +0200920ssize_t splice_from_pipe(struct pipe_inode_info *pipe, struct file *out,
921 loff_t *ppos, size_t len, unsigned int flags,
922 splice_actor *actor)
923{
924 ssize_t ret;
Jens Axboec66ab6f2007-06-12 21:17:17 +0200925 struct splice_desc sd = {
926 .total_len = len,
927 .flags = flags,
928 .pos = *ppos,
Jens Axboe6a14b902007-06-14 13:08:55 +0200929 .u.file = out,
Jens Axboec66ab6f2007-06-12 21:17:17 +0200930 };
Mark Fasheh6da61802006-10-17 18:43:07 +0200931
Miklos Szeredi61e0d472009-04-14 19:48:41 +0200932 pipe_lock(pipe);
Jens Axboec66ab6f2007-06-12 21:17:17 +0200933 ret = __splice_from_pipe(pipe, &sd, actor);
Miklos Szeredi61e0d472009-04-14 19:48:41 +0200934 pipe_unlock(pipe);
Mark Fasheh6da61802006-10-17 18:43:07 +0200935
936 return ret;
937}
938
939/**
Al Viro8d020762014-04-05 04:27:08 -0400940 * iter_file_splice_write - splice data from a pipe to a file
941 * @pipe: pipe info
942 * @out: file to write to
943 * @ppos: position in @out
944 * @len: number of bytes to splice
945 * @flags: splice modifier flags
946 *
947 * Description:
948 * Will either move or copy pages (determined by @flags options) from
949 * the given pipe inode to the given file.
950 * This one is ->write_iter-based.
951 *
952 */
953ssize_t
954iter_file_splice_write(struct pipe_inode_info *pipe, struct file *out,
955 loff_t *ppos, size_t len, unsigned int flags)
956{
957 struct splice_desc sd = {
958 .total_len = len,
959 .flags = flags,
960 .pos = *ppos,
961 .u.file = out,
962 };
963 int nbufs = pipe->buffers;
964 struct bio_vec *array = kcalloc(nbufs, sizeof(struct bio_vec),
965 GFP_KERNEL);
966 ssize_t ret;
967
968 if (unlikely(!array))
969 return -ENOMEM;
970
971 pipe_lock(pipe);
972
973 splice_from_pipe_begin(&sd);
974 while (sd.total_len) {
975 struct iov_iter from;
Al Viro8d020762014-04-05 04:27:08 -0400976 size_t left;
977 int n, idx;
978
979 ret = splice_from_pipe_next(pipe, &sd);
980 if (ret <= 0)
981 break;
982
983 if (unlikely(nbufs < pipe->buffers)) {
984 kfree(array);
985 nbufs = pipe->buffers;
986 array = kcalloc(nbufs, sizeof(struct bio_vec),
987 GFP_KERNEL);
988 if (!array) {
989 ret = -ENOMEM;
990 break;
991 }
992 }
993
994 /* build the vector */
995 left = sd.total_len;
996 for (n = 0, idx = pipe->curbuf; left && n < pipe->nrbufs; n++, idx++) {
997 struct pipe_buffer *buf = pipe->bufs + idx;
998 size_t this_len = buf->len;
999
1000 if (this_len > left)
1001 this_len = left;
1002
1003 if (idx == pipe->buffers - 1)
1004 idx = -1;
1005
1006 ret = buf->ops->confirm(pipe, buf);
1007 if (unlikely(ret)) {
1008 if (ret == -ENODATA)
1009 ret = 0;
1010 goto done;
1011 }
1012
1013 array[n].bv_page = buf->page;
1014 array[n].bv_len = this_len;
1015 array[n].bv_offset = buf->offset;
1016 left -= this_len;
1017 }
1018
Al Viro05afcb72015-01-23 01:08:07 -05001019 iov_iter_bvec(&from, ITER_BVEC | WRITE, array, n,
1020 sd.total_len - left);
Christoph Hellwigdbe4e192015-01-25 21:11:59 +01001021 ret = vfs_iter_write(out, &from, &sd.pos);
Al Viro8d020762014-04-05 04:27:08 -04001022 if (ret <= 0)
1023 break;
1024
1025 sd.num_spliced += ret;
1026 sd.total_len -= ret;
Christoph Hellwigdbe4e192015-01-25 21:11:59 +01001027 *ppos = sd.pos;
Al Viro8d020762014-04-05 04:27:08 -04001028
1029 /* dismiss the fully eaten buffers, adjust the partial one */
1030 while (ret) {
1031 struct pipe_buffer *buf = pipe->bufs + pipe->curbuf;
1032 if (ret >= buf->len) {
1033 const struct pipe_buf_operations *ops = buf->ops;
1034 ret -= buf->len;
1035 buf->len = 0;
1036 buf->ops = NULL;
1037 ops->release(pipe, buf);
1038 pipe->curbuf = (pipe->curbuf + 1) & (pipe->buffers - 1);
1039 pipe->nrbufs--;
1040 if (pipe->files)
1041 sd.need_wakeup = true;
1042 } else {
1043 buf->offset += ret;
1044 buf->len -= ret;
1045 ret = 0;
1046 }
1047 }
1048 }
1049done:
1050 kfree(array);
1051 splice_from_pipe_end(pipe, &sd);
1052
1053 pipe_unlock(pipe);
1054
1055 if (sd.num_spliced)
1056 ret = sd.num_spliced;
1057
1058 return ret;
1059}
1060
1061EXPORT_SYMBOL(iter_file_splice_write);
1062
Miklos Szeredib2858d72009-05-19 11:37:46 +02001063static int write_pipe_buf(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
1064 struct splice_desc *sd)
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +02001065{
Miklos Szeredib2858d72009-05-19 11:37:46 +02001066 int ret;
1067 void *data;
Al Viro06ae43f2013-03-20 13:19:30 -04001068 loff_t tmp = sd->pos;
Miklos Szeredib2858d72009-05-19 11:37:46 +02001069
Al Virofbb32752014-02-02 21:09:54 -05001070 data = kmap(buf->page);
Al Viro06ae43f2013-03-20 13:19:30 -04001071 ret = __kernel_write(sd->u.file, data + buf->offset, sd->len, &tmp);
Al Virofbb32752014-02-02 21:09:54 -05001072 kunmap(buf->page);
Miklos Szeredib2858d72009-05-19 11:37:46 +02001073
1074 return ret;
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +02001075}
1076
1077static ssize_t default_file_splice_write(struct pipe_inode_info *pipe,
1078 struct file *out, loff_t *ppos,
1079 size_t len, unsigned int flags)
1080{
Miklos Szeredib2858d72009-05-19 11:37:46 +02001081 ssize_t ret;
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +02001082
Miklos Szeredib2858d72009-05-19 11:37:46 +02001083 ret = splice_from_pipe(pipe, out, ppos, len, flags, write_pipe_buf);
1084 if (ret > 0)
1085 *ppos += ret;
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +02001086
Miklos Szeredib2858d72009-05-19 11:37:46 +02001087 return ret;
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +02001088}
1089
Jens Axboe83f91352006-04-02 23:05:09 +02001090/**
1091 * generic_splice_sendpage - splice data from a pipe to a socket
Jens Axboe932cc6d2007-06-21 13:10:21 +02001092 * @pipe: pipe to splice from
Jens Axboe83f91352006-04-02 23:05:09 +02001093 * @out: socket to write to
Jens Axboe932cc6d2007-06-21 13:10:21 +02001094 * @ppos: position in @out
Jens Axboe83f91352006-04-02 23:05:09 +02001095 * @len: number of bytes to splice
1096 * @flags: splice modifier flags
1097 *
Jens Axboe932cc6d2007-06-21 13:10:21 +02001098 * Description:
1099 * Will send @len bytes from the pipe to a network socket. No data copying
1100 * is involved.
Jens Axboe83f91352006-04-02 23:05:09 +02001101 *
1102 */
Ingo Molnar3a326a22006-04-10 15:18:35 +02001103ssize_t generic_splice_sendpage(struct pipe_inode_info *pipe, struct file *out,
Jens Axboecbb7e572006-04-11 14:57:50 +02001104 loff_t *ppos, size_t len, unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +02001105{
Jens Axboe00522fb2006-04-26 14:39:29 +02001106 return splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_sendpage);
Jens Axboe5274f052006-03-30 15:15:30 +02001107}
1108
Jens Axboe059a8f32006-04-02 23:06:05 +02001109EXPORT_SYMBOL(generic_splice_sendpage);
Jeff Garzika0f06782006-03-30 23:06:13 -05001110
Jens Axboe83f91352006-04-02 23:05:09 +02001111/*
1112 * Attempt to initiate a splice from pipe to file.
1113 */
Ingo Molnar3a326a22006-04-10 15:18:35 +02001114static long do_splice_from(struct pipe_inode_info *pipe, struct file *out,
Jens Axboecbb7e572006-04-11 14:57:50 +02001115 loff_t *ppos, size_t len, unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +02001116{
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +02001117 ssize_t (*splice_write)(struct pipe_inode_info *, struct file *,
1118 loff_t *, size_t, unsigned int);
Jens Axboe5274f052006-03-30 15:15:30 +02001119
Al Viro72c2d532013-09-22 16:27:52 -04001120 if (out->f_op->splice_write)
Changli Gaocc56f7d2009-11-04 09:09:52 +01001121 splice_write = out->f_op->splice_write;
1122 else
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +02001123 splice_write = default_file_splice_write;
1124
Al Viro500368f2013-05-23 20:07:11 -04001125 return splice_write(pipe, out, ppos, len, flags);
Jens Axboe5274f052006-03-30 15:15:30 +02001126}
1127
Jens Axboe83f91352006-04-02 23:05:09 +02001128/*
1129 * Attempt to initiate a splice from a file to a pipe.
1130 */
Jens Axboecbb7e572006-04-11 14:57:50 +02001131static long do_splice_to(struct file *in, loff_t *ppos,
1132 struct pipe_inode_info *pipe, size_t len,
1133 unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +02001134{
Miklos Szeredi68181732009-05-07 15:37:36 +02001135 ssize_t (*splice_read)(struct file *, loff_t *,
1136 struct pipe_inode_info *, size_t, unsigned int);
Jens Axboe5274f052006-03-30 15:15:30 +02001137 int ret;
1138
Jens Axboe49570e92006-04-11 13:56:09 +02001139 if (unlikely(!(in->f_mode & FMODE_READ)))
Jens Axboe5274f052006-03-30 15:15:30 +02001140 return -EBADF;
1141
Jens Axboecbb7e572006-04-11 14:57:50 +02001142 ret = rw_verify_area(READ, in, ppos, len);
Jens Axboe5274f052006-03-30 15:15:30 +02001143 if (unlikely(ret < 0))
1144 return ret;
1145
Al Viro03cc0782016-04-02 14:56:58 -04001146 if (unlikely(len > MAX_RW_COUNT))
1147 len = MAX_RW_COUNT;
1148
Al Viro72c2d532013-09-22 16:27:52 -04001149 if (in->f_op->splice_read)
Changli Gaocc56f7d2009-11-04 09:09:52 +01001150 splice_read = in->f_op->splice_read;
1151 else
Miklos Szeredi68181732009-05-07 15:37:36 +02001152 splice_read = default_file_splice_read;
1153
1154 return splice_read(in, ppos, pipe, len, flags);
Jens Axboe5274f052006-03-30 15:15:30 +02001155}
1156
Jens Axboe932cc6d2007-06-21 13:10:21 +02001157/**
1158 * splice_direct_to_actor - splices data directly between two non-pipes
1159 * @in: file to splice from
1160 * @sd: actor information on where to splice to
1161 * @actor: handles the data splicing
1162 *
1163 * Description:
1164 * This is a special case helper to splice directly between two
1165 * points, without requiring an explicit pipe. Internally an allocated
Randy Dunlap79685b82007-07-27 08:08:51 +02001166 * pipe is cached in the process, and reused during the lifetime of
Jens Axboe932cc6d2007-06-21 13:10:21 +02001167 * that process.
1168 *
Jens Axboec66ab6f2007-06-12 21:17:17 +02001169 */
1170ssize_t splice_direct_to_actor(struct file *in, struct splice_desc *sd,
1171 splice_direct_actor *actor)
Jens Axboeb92ce552006-04-11 13:52:07 +02001172{
1173 struct pipe_inode_info *pipe;
1174 long ret, bytes;
1175 umode_t i_mode;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001176 size_t len;
Christophe Leroy0ff28d92015-05-06 17:26:47 +02001177 int i, flags, more;
Jens Axboeb92ce552006-04-11 13:52:07 +02001178
1179 /*
1180 * We require the input being a regular file, as we don't want to
1181 * randomly drop data for eg socket -> socket splicing. Use the
1182 * piped splicing for that!
1183 */
Al Viro496ad9a2013-01-23 17:07:38 -05001184 i_mode = file_inode(in)->i_mode;
Jens Axboeb92ce552006-04-11 13:52:07 +02001185 if (unlikely(!S_ISREG(i_mode) && !S_ISBLK(i_mode)))
1186 return -EINVAL;
1187
1188 /*
1189 * neither in nor out is a pipe, setup an internal pipe attached to
1190 * 'out' and transfer the wanted data from 'in' to 'out' through that
1191 */
1192 pipe = current->splice_pipe;
Jens Axboe49570e92006-04-11 13:56:09 +02001193 if (unlikely(!pipe)) {
Al Viro7bee1302013-03-21 11:04:15 -04001194 pipe = alloc_pipe_info();
Jens Axboeb92ce552006-04-11 13:52:07 +02001195 if (!pipe)
1196 return -ENOMEM;
1197
1198 /*
1199 * We don't have an immediate reader, but we'll read the stuff
Jens Axboe00522fb2006-04-26 14:39:29 +02001200 * out of the pipe right after the splice_to_pipe(). So set
Jens Axboeb92ce552006-04-11 13:52:07 +02001201 * PIPE_READERS appropriately.
1202 */
1203 pipe->readers = 1;
1204
1205 current->splice_pipe = pipe;
1206 }
1207
1208 /*
Ingo Molnar73d62d82006-04-11 13:57:21 +02001209 * Do the splice.
Jens Axboeb92ce552006-04-11 13:52:07 +02001210 */
1211 ret = 0;
1212 bytes = 0;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001213 len = sd->total_len;
1214 flags = sd->flags;
1215
1216 /*
1217 * Don't block on output, we have to drain the direct pipe.
1218 */
1219 sd->flags &= ~SPLICE_F_NONBLOCK;
Christophe Leroy0ff28d92015-05-06 17:26:47 +02001220 more = sd->flags & SPLICE_F_MORE;
Jens Axboeb92ce552006-04-11 13:52:07 +02001221
1222 while (len) {
Jens Axboe51a92c02007-07-13 14:11:43 +02001223 size_t read_len;
Tom Zanussia82c53a2008-05-09 13:28:36 +02001224 loff_t pos = sd->pos, prev_pos = pos;
Jens Axboeb92ce552006-04-11 13:52:07 +02001225
Jens Axboebcd4f3a2007-07-16 14:41:49 +02001226 ret = do_splice_to(in, &pos, pipe, len, flags);
Jens Axboe51a92c02007-07-13 14:11:43 +02001227 if (unlikely(ret <= 0))
Jens Axboeb92ce552006-04-11 13:52:07 +02001228 goto out_release;
1229
1230 read_len = ret;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001231 sd->total_len = read_len;
Jens Axboeb92ce552006-04-11 13:52:07 +02001232
1233 /*
Christophe Leroy0ff28d92015-05-06 17:26:47 +02001234 * If more data is pending, set SPLICE_F_MORE
1235 * If this is the last data and SPLICE_F_MORE was not set
1236 * initially, clears it.
1237 */
1238 if (read_len < len)
1239 sd->flags |= SPLICE_F_MORE;
1240 else if (!more)
1241 sd->flags &= ~SPLICE_F_MORE;
1242 /*
Jens Axboeb92ce552006-04-11 13:52:07 +02001243 * NOTE: nonblocking mode only applies to the input. We
1244 * must not do the output in nonblocking mode as then we
1245 * could get stuck data in the internal pipe:
1246 */
Jens Axboec66ab6f2007-06-12 21:17:17 +02001247 ret = actor(pipe, sd);
Tom Zanussia82c53a2008-05-09 13:28:36 +02001248 if (unlikely(ret <= 0)) {
1249 sd->pos = prev_pos;
Jens Axboeb92ce552006-04-11 13:52:07 +02001250 goto out_release;
Tom Zanussia82c53a2008-05-09 13:28:36 +02001251 }
Jens Axboeb92ce552006-04-11 13:52:07 +02001252
1253 bytes += ret;
1254 len -= ret;
Jens Axboebcd4f3a2007-07-16 14:41:49 +02001255 sd->pos = pos;
Jens Axboeb92ce552006-04-11 13:52:07 +02001256
Tom Zanussia82c53a2008-05-09 13:28:36 +02001257 if (ret < read_len) {
1258 sd->pos = prev_pos + ret;
Jens Axboe51a92c02007-07-13 14:11:43 +02001259 goto out_release;
Tom Zanussia82c53a2008-05-09 13:28:36 +02001260 }
Jens Axboeb92ce552006-04-11 13:52:07 +02001261 }
1262
Jens Axboe9e971982008-01-29 21:05:57 +01001263done:
Jens Axboeb92ce552006-04-11 13:52:07 +02001264 pipe->nrbufs = pipe->curbuf = 0;
Jens Axboe80848702008-01-30 12:24:48 +01001265 file_accessed(in);
Jens Axboeb92ce552006-04-11 13:52:07 +02001266 return bytes;
1267
1268out_release:
1269 /*
1270 * If we did an incomplete transfer we must release
1271 * the pipe buffers in question:
1272 */
Jens Axboe35f3d142010-05-20 10:43:18 +02001273 for (i = 0; i < pipe->buffers; i++) {
Jens Axboeb92ce552006-04-11 13:52:07 +02001274 struct pipe_buffer *buf = pipe->bufs + i;
1275
1276 if (buf->ops) {
1277 buf->ops->release(pipe, buf);
1278 buf->ops = NULL;
1279 }
1280 }
Jens Axboeb92ce552006-04-11 13:52:07 +02001281
Jens Axboe9e971982008-01-29 21:05:57 +01001282 if (!bytes)
1283 bytes = ret;
Jens Axboeb92ce552006-04-11 13:52:07 +02001284
Jens Axboe9e971982008-01-29 21:05:57 +01001285 goto done;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001286}
1287EXPORT_SYMBOL(splice_direct_to_actor);
1288
1289static int direct_splice_actor(struct pipe_inode_info *pipe,
1290 struct splice_desc *sd)
1291{
Jens Axboe6a14b902007-06-14 13:08:55 +02001292 struct file *file = sd->u.file;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001293
Al Viro7995bd22013-06-20 18:58:36 +04001294 return do_splice_from(pipe, file, sd->opos, sd->total_len,
Changli Gao2cb4b052010-06-29 13:09:18 +02001295 sd->flags);
Jens Axboec66ab6f2007-06-12 21:17:17 +02001296}
1297
Jens Axboe932cc6d2007-06-21 13:10:21 +02001298/**
1299 * do_splice_direct - splices data directly between two files
1300 * @in: file to splice from
1301 * @ppos: input file offset
1302 * @out: file to splice to
Randy Dunlapacdb37c2013-06-22 19:44:08 -07001303 * @opos: output file offset
Jens Axboe932cc6d2007-06-21 13:10:21 +02001304 * @len: number of bytes to splice
1305 * @flags: splice modifier flags
1306 *
1307 * Description:
1308 * For use by do_sendfile(). splice can easily emulate sendfile, but
1309 * doing it in the application would incur an extra system call
1310 * (splice in + splice out, as compared to just sendfile()). So this helper
1311 * can splice directly through a process-private pipe.
1312 *
1313 */
Jens Axboec66ab6f2007-06-12 21:17:17 +02001314long do_splice_direct(struct file *in, loff_t *ppos, struct file *out,
Al Viro7995bd22013-06-20 18:58:36 +04001315 loff_t *opos, size_t len, unsigned int flags)
Jens Axboec66ab6f2007-06-12 21:17:17 +02001316{
1317 struct splice_desc sd = {
1318 .len = len,
1319 .total_len = len,
1320 .flags = flags,
1321 .pos = *ppos,
Jens Axboe6a14b902007-06-14 13:08:55 +02001322 .u.file = out,
Al Viro7995bd22013-06-20 18:58:36 +04001323 .opos = opos,
Jens Axboec66ab6f2007-06-12 21:17:17 +02001324 };
Jens Axboe51a92c02007-07-13 14:11:43 +02001325 long ret;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001326
Al Viro18c67cb2013-06-19 15:41:54 +04001327 if (unlikely(!(out->f_mode & FMODE_WRITE)))
1328 return -EBADF;
1329
1330 if (unlikely(out->f_flags & O_APPEND))
1331 return -EINVAL;
1332
1333 ret = rw_verify_area(WRITE, out, opos, len);
1334 if (unlikely(ret < 0))
1335 return ret;
1336
Jens Axboec66ab6f2007-06-12 21:17:17 +02001337 ret = splice_direct_to_actor(in, &sd, direct_splice_actor);
Jens Axboe51a92c02007-07-13 14:11:43 +02001338 if (ret > 0)
Tom Zanussia82c53a2008-05-09 13:28:36 +02001339 *ppos = sd.pos;
Jens Axboe51a92c02007-07-13 14:11:43 +02001340
Jens Axboec66ab6f2007-06-12 21:17:17 +02001341 return ret;
Jens Axboeb92ce552006-04-11 13:52:07 +02001342}
Miklos Szeredi1c118592014-10-24 00:14:35 +02001343EXPORT_SYMBOL(do_splice_direct);
Jens Axboeb92ce552006-04-11 13:52:07 +02001344
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001345static int splice_pipe_to_pipe(struct pipe_inode_info *ipipe,
1346 struct pipe_inode_info *opipe,
1347 size_t len, unsigned int flags);
Jens Axboeddac0d32006-11-04 12:49:32 +01001348
1349/*
Jens Axboe83f91352006-04-02 23:05:09 +02001350 * Determine where to splice to/from.
1351 */
Ingo Molnar529565d2006-04-10 15:18:58 +02001352static long do_splice(struct file *in, loff_t __user *off_in,
1353 struct file *out, loff_t __user *off_out,
1354 size_t len, unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +02001355{
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001356 struct pipe_inode_info *ipipe;
1357 struct pipe_inode_info *opipe;
Al Viro7995bd22013-06-20 18:58:36 +04001358 loff_t offset;
Jens Axboea4514eb2006-04-19 15:57:05 +02001359 long ret;
Jens Axboe5274f052006-03-30 15:15:30 +02001360
Linus Torvalds71993e62010-11-28 13:56:09 -08001361 ipipe = get_pipe_info(in);
1362 opipe = get_pipe_info(out);
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001363
1364 if (ipipe && opipe) {
1365 if (off_in || off_out)
1366 return -ESPIPE;
1367
1368 if (!(in->f_mode & FMODE_READ))
1369 return -EBADF;
1370
1371 if (!(out->f_mode & FMODE_WRITE))
1372 return -EBADF;
1373
1374 /* Splicing to self would be fun, but... */
1375 if (ipipe == opipe)
1376 return -EINVAL;
1377
1378 return splice_pipe_to_pipe(ipipe, opipe, len, flags);
1379 }
1380
1381 if (ipipe) {
Ingo Molnar529565d2006-04-10 15:18:58 +02001382 if (off_in)
1383 return -ESPIPE;
Jens Axboeb92ce552006-04-11 13:52:07 +02001384 if (off_out) {
Changli Gao19c9a492010-06-29 13:10:36 +02001385 if (!(out->f_mode & FMODE_PWRITE))
Jens Axboeb92ce552006-04-11 13:52:07 +02001386 return -EINVAL;
Jens Axboecbb7e572006-04-11 14:57:50 +02001387 if (copy_from_user(&offset, off_out, sizeof(loff_t)))
Jens Axboeb92ce552006-04-11 13:52:07 +02001388 return -EFAULT;
Al Viro7995bd22013-06-20 18:58:36 +04001389 } else {
1390 offset = out->f_pos;
1391 }
Ingo Molnar529565d2006-04-10 15:18:58 +02001392
Al Viro18c67cb2013-06-19 15:41:54 +04001393 if (unlikely(!(out->f_mode & FMODE_WRITE)))
1394 return -EBADF;
1395
1396 if (unlikely(out->f_flags & O_APPEND))
1397 return -EINVAL;
1398
1399 ret = rw_verify_area(WRITE, out, &offset, len);
1400 if (unlikely(ret < 0))
1401 return ret;
1402
Al Viro500368f2013-05-23 20:07:11 -04001403 file_start_write(out);
Al Viro7995bd22013-06-20 18:58:36 +04001404 ret = do_splice_from(ipipe, out, &offset, len, flags);
Al Viro500368f2013-05-23 20:07:11 -04001405 file_end_write(out);
Jens Axboea4514eb2006-04-19 15:57:05 +02001406
Al Viro7995bd22013-06-20 18:58:36 +04001407 if (!off_out)
1408 out->f_pos = offset;
1409 else if (copy_to_user(off_out, &offset, sizeof(loff_t)))
Jens Axboea4514eb2006-04-19 15:57:05 +02001410 ret = -EFAULT;
1411
1412 return ret;
Ingo Molnar529565d2006-04-10 15:18:58 +02001413 }
Jens Axboe5274f052006-03-30 15:15:30 +02001414
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001415 if (opipe) {
Ingo Molnar529565d2006-04-10 15:18:58 +02001416 if (off_out)
1417 return -ESPIPE;
Jens Axboeb92ce552006-04-11 13:52:07 +02001418 if (off_in) {
Changli Gao19c9a492010-06-29 13:10:36 +02001419 if (!(in->f_mode & FMODE_PREAD))
Jens Axboeb92ce552006-04-11 13:52:07 +02001420 return -EINVAL;
Jens Axboecbb7e572006-04-11 14:57:50 +02001421 if (copy_from_user(&offset, off_in, sizeof(loff_t)))
Jens Axboeb92ce552006-04-11 13:52:07 +02001422 return -EFAULT;
Al Viro7995bd22013-06-20 18:58:36 +04001423 } else {
1424 offset = in->f_pos;
1425 }
Ingo Molnar529565d2006-04-10 15:18:58 +02001426
Al Viro7995bd22013-06-20 18:58:36 +04001427 ret = do_splice_to(in, &offset, opipe, len, flags);
Jens Axboea4514eb2006-04-19 15:57:05 +02001428
Al Viro7995bd22013-06-20 18:58:36 +04001429 if (!off_in)
1430 in->f_pos = offset;
1431 else if (copy_to_user(off_in, &offset, sizeof(loff_t)))
Jens Axboea4514eb2006-04-19 15:57:05 +02001432 ret = -EFAULT;
1433
1434 return ret;
Ingo Molnar529565d2006-04-10 15:18:58 +02001435 }
Jens Axboe5274f052006-03-30 15:15:30 +02001436
1437 return -EINVAL;
1438}
1439
Jens Axboe912d35f2006-04-26 10:59:21 +02001440/*
1441 * Map an iov into an array of pages and offset/length tupples. With the
1442 * partial_page structure, we can map several non-contiguous ranges into
1443 * our ones pages[] map instead of splitting that operation into pieces.
1444 * Could easily be exported as a generic helper for other users, in which
1445 * case one would probably want to add a 'max_nr_pages' parameter as well.
1446 */
1447static int get_iovec_page_array(const struct iovec __user *iov,
1448 unsigned int nr_vecs, struct page **pages,
Eric Dumazetbd1a68b2012-04-04 11:08:51 +02001449 struct partial_page *partial, bool aligned,
Jens Axboe35f3d142010-05-20 10:43:18 +02001450 unsigned int pipe_buffers)
Jens Axboe912d35f2006-04-26 10:59:21 +02001451{
1452 int buffers = 0, error = 0;
1453
Jens Axboe912d35f2006-04-26 10:59:21 +02001454 while (nr_vecs) {
1455 unsigned long off, npages;
Linus Torvalds75723952007-10-01 13:17:28 -07001456 struct iovec entry;
Jens Axboe912d35f2006-04-26 10:59:21 +02001457 void __user *base;
1458 size_t len;
1459 int i;
1460
Linus Torvalds75723952007-10-01 13:17:28 -07001461 error = -EFAULT;
Nick Pigginbc40d732008-07-25 19:45:26 -07001462 if (copy_from_user(&entry, iov, sizeof(entry)))
Jens Axboe912d35f2006-04-26 10:59:21 +02001463 break;
Linus Torvalds75723952007-10-01 13:17:28 -07001464
1465 base = entry.iov_base;
1466 len = entry.iov_len;
Jens Axboe912d35f2006-04-26 10:59:21 +02001467
1468 /*
1469 * Sanity check this iovec. 0 read succeeds.
1470 */
Linus Torvalds75723952007-10-01 13:17:28 -07001471 error = 0;
Jens Axboe912d35f2006-04-26 10:59:21 +02001472 if (unlikely(!len))
1473 break;
1474 error = -EFAULT;
Bastian Blank712a30e2008-02-10 16:47:57 +02001475 if (!access_ok(VERIFY_READ, base, len))
Jens Axboe912d35f2006-04-26 10:59:21 +02001476 break;
1477
1478 /*
1479 * Get this base offset and number of pages, then map
1480 * in the user pages.
1481 */
1482 off = (unsigned long) base & ~PAGE_MASK;
Jens Axboe7afa6fd2006-05-01 20:02:33 +02001483
1484 /*
1485 * If asked for alignment, the offset must be zero and the
1486 * length a multiple of the PAGE_SIZE.
1487 */
1488 error = -EINVAL;
1489 if (aligned && (off || len & ~PAGE_MASK))
1490 break;
1491
Jens Axboe912d35f2006-04-26 10:59:21 +02001492 npages = (off + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
Jens Axboe35f3d142010-05-20 10:43:18 +02001493 if (npages > pipe_buffers - buffers)
1494 npages = pipe_buffers - buffers;
Jens Axboe912d35f2006-04-26 10:59:21 +02001495
Nick Pigginbc40d732008-07-25 19:45:26 -07001496 error = get_user_pages_fast((unsigned long)base, npages,
1497 0, &pages[buffers]);
Jens Axboe912d35f2006-04-26 10:59:21 +02001498
1499 if (unlikely(error <= 0))
1500 break;
1501
1502 /*
1503 * Fill this contiguous range into the partial page map.
1504 */
1505 for (i = 0; i < error; i++) {
Jens Axboe75914892006-05-02 12:57:18 +02001506 const int plen = min_t(size_t, len, PAGE_SIZE - off);
Jens Axboe912d35f2006-04-26 10:59:21 +02001507
1508 partial[buffers].offset = off;
1509 partial[buffers].len = plen;
1510
1511 off = 0;
1512 len -= plen;
1513 buffers++;
1514 }
1515
1516 /*
1517 * We didn't complete this iov, stop here since it probably
1518 * means we have to move some of this into a pipe to
1519 * be able to continue.
1520 */
1521 if (len)
1522 break;
1523
1524 /*
1525 * Don't continue if we mapped fewer pages than we asked for,
1526 * or if we mapped the max number of pages that we have
1527 * room for.
1528 */
Jens Axboe35f3d142010-05-20 10:43:18 +02001529 if (error < npages || buffers == pipe_buffers)
Jens Axboe912d35f2006-04-26 10:59:21 +02001530 break;
1531
1532 nr_vecs--;
1533 iov++;
1534 }
1535
Jens Axboe912d35f2006-04-26 10:59:21 +02001536 if (buffers)
1537 return buffers;
1538
1539 return error;
1540}
1541
Jens Axboe6a14b902007-06-14 13:08:55 +02001542static int pipe_to_user(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
1543 struct splice_desc *sd)
1544{
Al Viro6130f532014-02-03 18:19:51 -05001545 int n = copy_page_to_iter(buf->page, buf->offset, sd->len, sd->u.data);
1546 return n == sd->len ? n : -EFAULT;
Jens Axboe6a14b902007-06-14 13:08:55 +02001547}
1548
1549/*
1550 * For lack of a better implementation, implement vmsplice() to userspace
1551 * as a simple copy of the pipes pages to the user iov.
1552 */
Al Viro6130f532014-02-03 18:19:51 -05001553static long vmsplice_to_user(struct file *file, const struct iovec __user *uiov,
Jens Axboe6a14b902007-06-14 13:08:55 +02001554 unsigned long nr_segs, unsigned int flags)
1555{
1556 struct pipe_inode_info *pipe;
1557 struct splice_desc sd;
Jens Axboe6a14b902007-06-14 13:08:55 +02001558 long ret;
Al Viro6130f532014-02-03 18:19:51 -05001559 struct iovec iovstack[UIO_FASTIOV];
1560 struct iovec *iov = iovstack;
1561 struct iov_iter iter;
Jens Axboe6a14b902007-06-14 13:08:55 +02001562
Linus Torvalds71993e62010-11-28 13:56:09 -08001563 pipe = get_pipe_info(file);
Jens Axboe6a14b902007-06-14 13:08:55 +02001564 if (!pipe)
1565 return -EBADF;
1566
Al Viro345995f2015-03-21 19:17:55 -04001567 ret = import_iovec(READ, uiov, nr_segs,
1568 ARRAY_SIZE(iovstack), &iov, &iter);
1569 if (ret < 0)
1570 return ret;
Al Viro6130f532014-02-03 18:19:51 -05001571
Al Viro345995f2015-03-21 19:17:55 -04001572 sd.total_len = iov_iter_count(&iter);
Al Viro6130f532014-02-03 18:19:51 -05001573 sd.len = 0;
Al Viro6130f532014-02-03 18:19:51 -05001574 sd.flags = flags;
1575 sd.u.data = &iter;
1576 sd.pos = 0;
1577
Al Viro345995f2015-03-21 19:17:55 -04001578 if (sd.total_len) {
1579 pipe_lock(pipe);
1580 ret = __splice_from_pipe(pipe, &sd, pipe_to_user);
1581 pipe_unlock(pipe);
1582 }
Jens Axboe6a14b902007-06-14 13:08:55 +02001583
Al Viro345995f2015-03-21 19:17:55 -04001584 kfree(iov);
Jens Axboe6a14b902007-06-14 13:08:55 +02001585 return ret;
1586}
1587
Jens Axboe912d35f2006-04-26 10:59:21 +02001588/*
1589 * vmsplice splices a user address range into a pipe. It can be thought of
1590 * as splice-from-memory, where the regular splice is splice-from-file (or
1591 * to file). In both cases the output is a pipe, naturally.
Jens Axboe912d35f2006-04-26 10:59:21 +02001592 */
Jens Axboe6a14b902007-06-14 13:08:55 +02001593static long vmsplice_to_pipe(struct file *file, const struct iovec __user *iov,
1594 unsigned long nr_segs, unsigned int flags)
Jens Axboe912d35f2006-04-26 10:59:21 +02001595{
Jens Axboeddac0d32006-11-04 12:49:32 +01001596 struct pipe_inode_info *pipe;
Jens Axboe35f3d142010-05-20 10:43:18 +02001597 struct page *pages[PIPE_DEF_BUFFERS];
1598 struct partial_page partial[PIPE_DEF_BUFFERS];
Jens Axboe912d35f2006-04-26 10:59:21 +02001599 struct splice_pipe_desc spd = {
1600 .pages = pages,
1601 .partial = partial,
Eric Dumazet047fe362012-06-12 15:24:40 +02001602 .nr_pages_max = PIPE_DEF_BUFFERS,
Jens Axboe912d35f2006-04-26 10:59:21 +02001603 .flags = flags,
1604 .ops = &user_page_pipe_buf_ops,
Jens Axboebbdfc2f2007-11-06 23:29:47 -08001605 .spd_release = spd_release_page,
Jens Axboe912d35f2006-04-26 10:59:21 +02001606 };
Jens Axboe35f3d142010-05-20 10:43:18 +02001607 long ret;
Jens Axboe912d35f2006-04-26 10:59:21 +02001608
Linus Torvalds71993e62010-11-28 13:56:09 -08001609 pipe = get_pipe_info(file);
Jens Axboeddac0d32006-11-04 12:49:32 +01001610 if (!pipe)
Jens Axboe912d35f2006-04-26 10:59:21 +02001611 return -EBADF;
Jens Axboe912d35f2006-04-26 10:59:21 +02001612
Jens Axboe35f3d142010-05-20 10:43:18 +02001613 if (splice_grow_spd(pipe, &spd))
1614 return -ENOMEM;
Jens Axboe912d35f2006-04-26 10:59:21 +02001615
Jens Axboe35f3d142010-05-20 10:43:18 +02001616 spd.nr_pages = get_iovec_page_array(iov, nr_segs, spd.pages,
Eric Dumazetbd1a68b2012-04-04 11:08:51 +02001617 spd.partial, false,
Eric Dumazet047fe362012-06-12 15:24:40 +02001618 spd.nr_pages_max);
Jens Axboe35f3d142010-05-20 10:43:18 +02001619 if (spd.nr_pages <= 0)
1620 ret = spd.nr_pages;
1621 else
1622 ret = splice_to_pipe(pipe, &spd);
1623
Eric Dumazet047fe362012-06-12 15:24:40 +02001624 splice_shrink_spd(&spd);
Jens Axboe35f3d142010-05-20 10:43:18 +02001625 return ret;
Jens Axboe912d35f2006-04-26 10:59:21 +02001626}
1627
Jens Axboe6a14b902007-06-14 13:08:55 +02001628/*
1629 * Note that vmsplice only really supports true splicing _from_ user memory
1630 * to a pipe, not the other way around. Splicing from user memory is a simple
1631 * operation that can be supported without any funky alignment restrictions
1632 * or nasty vm tricks. We simply map in the user memory and fill them into
1633 * a pipe. The reverse isn't quite as easy, though. There are two possible
1634 * solutions for that:
1635 *
1636 * - memcpy() the data internally, at which point we might as well just
1637 * do a regular read() on the buffer anyway.
1638 * - Lots of nasty vm tricks, that are neither fast nor flexible (it
1639 * has restriction limitations on both ends of the pipe).
1640 *
1641 * Currently we punt and implement it as a normal copy, see pipe_to_user().
1642 *
1643 */
Heiko Carstens836f92a2009-01-14 14:14:33 +01001644SYSCALL_DEFINE4(vmsplice, int, fd, const struct iovec __user *, iov,
1645 unsigned long, nr_segs, unsigned int, flags)
Jens Axboe912d35f2006-04-26 10:59:21 +02001646{
Al Viro2903ff02012-08-28 12:52:22 -04001647 struct fd f;
Jens Axboe912d35f2006-04-26 10:59:21 +02001648 long error;
Jens Axboe912d35f2006-04-26 10:59:21 +02001649
Jens Axboe6a14b902007-06-14 13:08:55 +02001650 if (unlikely(nr_segs > UIO_MAXIOV))
1651 return -EINVAL;
1652 else if (unlikely(!nr_segs))
1653 return 0;
1654
Jens Axboe912d35f2006-04-26 10:59:21 +02001655 error = -EBADF;
Al Viro2903ff02012-08-28 12:52:22 -04001656 f = fdget(fd);
1657 if (f.file) {
1658 if (f.file->f_mode & FMODE_WRITE)
1659 error = vmsplice_to_pipe(f.file, iov, nr_segs, flags);
1660 else if (f.file->f_mode & FMODE_READ)
1661 error = vmsplice_to_user(f.file, iov, nr_segs, flags);
Jens Axboe912d35f2006-04-26 10:59:21 +02001662
Al Viro2903ff02012-08-28 12:52:22 -04001663 fdput(f);
Jens Axboe912d35f2006-04-26 10:59:21 +02001664 }
1665
1666 return error;
1667}
1668
Al Viro76b021d2013-03-02 10:19:56 -05001669#ifdef CONFIG_COMPAT
1670COMPAT_SYSCALL_DEFINE4(vmsplice, int, fd, const struct compat_iovec __user *, iov32,
1671 unsigned int, nr_segs, unsigned int, flags)
1672{
1673 unsigned i;
1674 struct iovec __user *iov;
1675 if (nr_segs > UIO_MAXIOV)
1676 return -EINVAL;
1677 iov = compat_alloc_user_space(nr_segs * sizeof(struct iovec));
1678 for (i = 0; i < nr_segs; i++) {
1679 struct compat_iovec v;
1680 if (get_user(v.iov_base, &iov32[i].iov_base) ||
1681 get_user(v.iov_len, &iov32[i].iov_len) ||
1682 put_user(compat_ptr(v.iov_base), &iov[i].iov_base) ||
1683 put_user(v.iov_len, &iov[i].iov_len))
1684 return -EFAULT;
1685 }
1686 return sys_vmsplice(fd, iov, nr_segs, flags);
1687}
1688#endif
1689
Heiko Carstens836f92a2009-01-14 14:14:33 +01001690SYSCALL_DEFINE6(splice, int, fd_in, loff_t __user *, off_in,
1691 int, fd_out, loff_t __user *, off_out,
1692 size_t, len, unsigned int, flags)
Jens Axboe5274f052006-03-30 15:15:30 +02001693{
Al Viro2903ff02012-08-28 12:52:22 -04001694 struct fd in, out;
Jens Axboe5274f052006-03-30 15:15:30 +02001695 long error;
Jens Axboe5274f052006-03-30 15:15:30 +02001696
1697 if (unlikely(!len))
1698 return 0;
1699
1700 error = -EBADF;
Al Viro2903ff02012-08-28 12:52:22 -04001701 in = fdget(fd_in);
1702 if (in.file) {
1703 if (in.file->f_mode & FMODE_READ) {
1704 out = fdget(fd_out);
1705 if (out.file) {
1706 if (out.file->f_mode & FMODE_WRITE)
1707 error = do_splice(in.file, off_in,
1708 out.file, off_out,
Ingo Molnar529565d2006-04-10 15:18:58 +02001709 len, flags);
Al Viro2903ff02012-08-28 12:52:22 -04001710 fdput(out);
Jens Axboe5274f052006-03-30 15:15:30 +02001711 }
1712 }
Al Viro2903ff02012-08-28 12:52:22 -04001713 fdput(in);
Jens Axboe5274f052006-03-30 15:15:30 +02001714 }
Jens Axboe5274f052006-03-30 15:15:30 +02001715 return error;
1716}
Jens Axboe70524492006-04-11 15:51:17 +02001717
1718/*
Jens Axboeaadd06e2006-07-10 11:00:01 +02001719 * Make sure there's data to read. Wait for input if we can, otherwise
1720 * return an appropriate error.
1721 */
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001722static int ipipe_prep(struct pipe_inode_info *pipe, unsigned int flags)
Jens Axboeaadd06e2006-07-10 11:00:01 +02001723{
1724 int ret;
1725
1726 /*
1727 * Check ->nrbufs without the inode lock first. This function
1728 * is speculative anyways, so missing one is ok.
1729 */
1730 if (pipe->nrbufs)
1731 return 0;
1732
1733 ret = 0;
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001734 pipe_lock(pipe);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001735
1736 while (!pipe->nrbufs) {
1737 if (signal_pending(current)) {
1738 ret = -ERESTARTSYS;
1739 break;
1740 }
1741 if (!pipe->writers)
1742 break;
1743 if (!pipe->waiting_writers) {
1744 if (flags & SPLICE_F_NONBLOCK) {
1745 ret = -EAGAIN;
1746 break;
1747 }
1748 }
1749 pipe_wait(pipe);
1750 }
1751
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001752 pipe_unlock(pipe);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001753 return ret;
1754}
1755
1756/*
1757 * Make sure there's writeable room. Wait for room if we can, otherwise
1758 * return an appropriate error.
1759 */
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001760static int opipe_prep(struct pipe_inode_info *pipe, unsigned int flags)
Jens Axboeaadd06e2006-07-10 11:00:01 +02001761{
1762 int ret;
1763
1764 /*
1765 * Check ->nrbufs without the inode lock first. This function
1766 * is speculative anyways, so missing one is ok.
1767 */
Jens Axboe35f3d142010-05-20 10:43:18 +02001768 if (pipe->nrbufs < pipe->buffers)
Jens Axboeaadd06e2006-07-10 11:00:01 +02001769 return 0;
1770
1771 ret = 0;
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001772 pipe_lock(pipe);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001773
Jens Axboe35f3d142010-05-20 10:43:18 +02001774 while (pipe->nrbufs >= pipe->buffers) {
Jens Axboeaadd06e2006-07-10 11:00:01 +02001775 if (!pipe->readers) {
1776 send_sig(SIGPIPE, current, 0);
1777 ret = -EPIPE;
1778 break;
1779 }
1780 if (flags & SPLICE_F_NONBLOCK) {
1781 ret = -EAGAIN;
1782 break;
1783 }
1784 if (signal_pending(current)) {
1785 ret = -ERESTARTSYS;
1786 break;
1787 }
1788 pipe->waiting_writers++;
1789 pipe_wait(pipe);
1790 pipe->waiting_writers--;
1791 }
1792
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001793 pipe_unlock(pipe);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001794 return ret;
1795}
1796
1797/*
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001798 * Splice contents of ipipe to opipe.
1799 */
1800static int splice_pipe_to_pipe(struct pipe_inode_info *ipipe,
1801 struct pipe_inode_info *opipe,
1802 size_t len, unsigned int flags)
1803{
1804 struct pipe_buffer *ibuf, *obuf;
1805 int ret = 0, nbuf;
1806 bool input_wakeup = false;
1807
1808
1809retry:
1810 ret = ipipe_prep(ipipe, flags);
1811 if (ret)
1812 return ret;
1813
1814 ret = opipe_prep(opipe, flags);
1815 if (ret)
1816 return ret;
1817
1818 /*
1819 * Potential ABBA deadlock, work around it by ordering lock
1820 * grabbing by pipe info address. Otherwise two different processes
1821 * could deadlock (one doing tee from A -> B, the other from B -> A).
1822 */
1823 pipe_double_lock(ipipe, opipe);
1824
1825 do {
1826 if (!opipe->readers) {
1827 send_sig(SIGPIPE, current, 0);
1828 if (!ret)
1829 ret = -EPIPE;
1830 break;
1831 }
1832
1833 if (!ipipe->nrbufs && !ipipe->writers)
1834 break;
1835
1836 /*
1837 * Cannot make any progress, because either the input
1838 * pipe is empty or the output pipe is full.
1839 */
Jens Axboe35f3d142010-05-20 10:43:18 +02001840 if (!ipipe->nrbufs || opipe->nrbufs >= opipe->buffers) {
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001841 /* Already processed some buffers, break */
1842 if (ret)
1843 break;
1844
1845 if (flags & SPLICE_F_NONBLOCK) {
1846 ret = -EAGAIN;
1847 break;
1848 }
1849
1850 /*
1851 * We raced with another reader/writer and haven't
1852 * managed to process any buffers. A zero return
1853 * value means EOF, so retry instead.
1854 */
1855 pipe_unlock(ipipe);
1856 pipe_unlock(opipe);
1857 goto retry;
1858 }
1859
1860 ibuf = ipipe->bufs + ipipe->curbuf;
Jens Axboe35f3d142010-05-20 10:43:18 +02001861 nbuf = (opipe->curbuf + opipe->nrbufs) & (opipe->buffers - 1);
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001862 obuf = opipe->bufs + nbuf;
1863
1864 if (len >= ibuf->len) {
1865 /*
1866 * Simply move the whole buffer from ipipe to opipe
1867 */
1868 *obuf = *ibuf;
1869 ibuf->ops = NULL;
1870 opipe->nrbufs++;
Jens Axboe35f3d142010-05-20 10:43:18 +02001871 ipipe->curbuf = (ipipe->curbuf + 1) & (ipipe->buffers - 1);
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001872 ipipe->nrbufs--;
1873 input_wakeup = true;
1874 } else {
1875 /*
1876 * Get a reference to this pipe buffer,
1877 * so we can copy the contents over.
1878 */
1879 ibuf->ops->get(ipipe, ibuf);
1880 *obuf = *ibuf;
1881
1882 /*
1883 * Don't inherit the gift flag, we need to
1884 * prevent multiple steals of this page.
1885 */
1886 obuf->flags &= ~PIPE_BUF_FLAG_GIFT;
1887
1888 obuf->len = len;
1889 opipe->nrbufs++;
1890 ibuf->offset += obuf->len;
1891 ibuf->len -= obuf->len;
1892 }
1893 ret += obuf->len;
1894 len -= obuf->len;
1895 } while (len);
1896
1897 pipe_unlock(ipipe);
1898 pipe_unlock(opipe);
1899
1900 /*
1901 * If we put data in the output pipe, wakeup any potential readers.
1902 */
Namhyung Kim825cdcb2011-05-23 19:58:53 +02001903 if (ret > 0)
1904 wakeup_pipe_readers(opipe);
1905
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001906 if (input_wakeup)
1907 wakeup_pipe_writers(ipipe);
1908
1909 return ret;
1910}
1911
1912/*
Jens Axboe70524492006-04-11 15:51:17 +02001913 * Link contents of ipipe to opipe.
1914 */
1915static int link_pipe(struct pipe_inode_info *ipipe,
1916 struct pipe_inode_info *opipe,
1917 size_t len, unsigned int flags)
1918{
1919 struct pipe_buffer *ibuf, *obuf;
Jens Axboeaadd06e2006-07-10 11:00:01 +02001920 int ret = 0, i = 0, nbuf;
Jens Axboe70524492006-04-11 15:51:17 +02001921
1922 /*
1923 * Potential ABBA deadlock, work around it by ordering lock
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001924 * grabbing by pipe info address. Otherwise two different processes
Jens Axboe70524492006-04-11 15:51:17 +02001925 * could deadlock (one doing tee from A -> B, the other from B -> A).
1926 */
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001927 pipe_double_lock(ipipe, opipe);
Jens Axboe70524492006-04-11 15:51:17 +02001928
Jens Axboeaadd06e2006-07-10 11:00:01 +02001929 do {
Jens Axboe70524492006-04-11 15:51:17 +02001930 if (!opipe->readers) {
1931 send_sig(SIGPIPE, current, 0);
1932 if (!ret)
1933 ret = -EPIPE;
1934 break;
1935 }
Jens Axboe70524492006-04-11 15:51:17 +02001936
1937 /*
Jens Axboeaadd06e2006-07-10 11:00:01 +02001938 * If we have iterated all input buffers or ran out of
1939 * output room, break.
Jens Axboe70524492006-04-11 15:51:17 +02001940 */
Jens Axboe35f3d142010-05-20 10:43:18 +02001941 if (i >= ipipe->nrbufs || opipe->nrbufs >= opipe->buffers)
Jens Axboe70524492006-04-11 15:51:17 +02001942 break;
Jens Axboeaadd06e2006-07-10 11:00:01 +02001943
Jens Axboe35f3d142010-05-20 10:43:18 +02001944 ibuf = ipipe->bufs + ((ipipe->curbuf + i) & (ipipe->buffers-1));
1945 nbuf = (opipe->curbuf + opipe->nrbufs) & (opipe->buffers - 1);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001946
Jens Axboe2a27250e2006-04-19 15:56:40 +02001947 /*
Jens Axboeaadd06e2006-07-10 11:00:01 +02001948 * Get a reference to this pipe buffer,
1949 * so we can copy the contents over.
Jens Axboe2a27250e2006-04-19 15:56:40 +02001950 */
Jens Axboeaadd06e2006-07-10 11:00:01 +02001951 ibuf->ops->get(ipipe, ibuf);
Jens Axboe70524492006-04-11 15:51:17 +02001952
Jens Axboeaadd06e2006-07-10 11:00:01 +02001953 obuf = opipe->bufs + nbuf;
1954 *obuf = *ibuf;
Jens Axboe70524492006-04-11 15:51:17 +02001955
Jens Axboeaadd06e2006-07-10 11:00:01 +02001956 /*
1957 * Don't inherit the gift flag, we need to
1958 * prevent multiple steals of this page.
1959 */
1960 obuf->flags &= ~PIPE_BUF_FLAG_GIFT;
1961
1962 if (obuf->len > len)
1963 obuf->len = len;
1964
1965 opipe->nrbufs++;
1966 ret += obuf->len;
1967 len -= obuf->len;
1968 i++;
1969 } while (len);
Jens Axboe70524492006-04-11 15:51:17 +02001970
Jens Axboe02cf01a2008-02-20 10:34:51 +01001971 /*
1972 * return EAGAIN if we have the potential of some data in the
1973 * future, otherwise just return 0
1974 */
1975 if (!ret && ipipe->waiting_writers && (flags & SPLICE_F_NONBLOCK))
1976 ret = -EAGAIN;
1977
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001978 pipe_unlock(ipipe);
1979 pipe_unlock(opipe);
Jens Axboe70524492006-04-11 15:51:17 +02001980
Jens Axboeaadd06e2006-07-10 11:00:01 +02001981 /*
1982 * If we put data in the output pipe, wakeup any potential readers.
1983 */
Namhyung Kim825cdcb2011-05-23 19:58:53 +02001984 if (ret > 0)
1985 wakeup_pipe_readers(opipe);
Jens Axboe70524492006-04-11 15:51:17 +02001986
1987 return ret;
1988}
1989
1990/*
1991 * This is a tee(1) implementation that works on pipes. It doesn't copy
1992 * any data, it simply references the 'in' pages on the 'out' pipe.
1993 * The 'flags' used are the SPLICE_F_* variants, currently the only
1994 * applicable one is SPLICE_F_NONBLOCK.
1995 */
1996static long do_tee(struct file *in, struct file *out, size_t len,
1997 unsigned int flags)
1998{
Linus Torvalds71993e62010-11-28 13:56:09 -08001999 struct pipe_inode_info *ipipe = get_pipe_info(in);
2000 struct pipe_inode_info *opipe = get_pipe_info(out);
Jens Axboeaadd06e2006-07-10 11:00:01 +02002001 int ret = -EINVAL;
Jens Axboe70524492006-04-11 15:51:17 +02002002
2003 /*
Jens Axboeaadd06e2006-07-10 11:00:01 +02002004 * Duplicate the contents of ipipe to opipe without actually
2005 * copying the data.
Jens Axboe70524492006-04-11 15:51:17 +02002006 */
Jens Axboeaadd06e2006-07-10 11:00:01 +02002007 if (ipipe && opipe && ipipe != opipe) {
2008 /*
2009 * Keep going, unless we encounter an error. The ipipe/opipe
2010 * ordering doesn't really matter.
2011 */
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02002012 ret = ipipe_prep(ipipe, flags);
Jens Axboeaadd06e2006-07-10 11:00:01 +02002013 if (!ret) {
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02002014 ret = opipe_prep(opipe, flags);
Jens Axboe02cf01a2008-02-20 10:34:51 +01002015 if (!ret)
Jens Axboeaadd06e2006-07-10 11:00:01 +02002016 ret = link_pipe(ipipe, opipe, len, flags);
Jens Axboeaadd06e2006-07-10 11:00:01 +02002017 }
2018 }
Jens Axboe70524492006-04-11 15:51:17 +02002019
Jens Axboeaadd06e2006-07-10 11:00:01 +02002020 return ret;
Jens Axboe70524492006-04-11 15:51:17 +02002021}
2022
Heiko Carstens836f92a2009-01-14 14:14:33 +01002023SYSCALL_DEFINE4(tee, int, fdin, int, fdout, size_t, len, unsigned int, flags)
Jens Axboe70524492006-04-11 15:51:17 +02002024{
Al Viro2903ff02012-08-28 12:52:22 -04002025 struct fd in;
2026 int error;
Jens Axboe70524492006-04-11 15:51:17 +02002027
2028 if (unlikely(!len))
2029 return 0;
2030
2031 error = -EBADF;
Al Viro2903ff02012-08-28 12:52:22 -04002032 in = fdget(fdin);
2033 if (in.file) {
2034 if (in.file->f_mode & FMODE_READ) {
2035 struct fd out = fdget(fdout);
2036 if (out.file) {
2037 if (out.file->f_mode & FMODE_WRITE)
2038 error = do_tee(in.file, out.file,
2039 len, flags);
2040 fdput(out);
Jens Axboe70524492006-04-11 15:51:17 +02002041 }
2042 }
Al Viro2903ff02012-08-28 12:52:22 -04002043 fdput(in);
Jens Axboe70524492006-04-11 15:51:17 +02002044 }
2045
2046 return error;
2047}