blob: 801c21cd77fe4f1d344aed6fa7b28369972d82b2 [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{
91 page_cache_release(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
188 ret = 0;
189 do_wakeup = 0;
Jens Axboe912d35f2006-04-26 10:59:21 +0200190 page_nr = 0;
Jens Axboe5274f052006-03-30 15:15:30 +0200191
Miklos Szeredi61e0d472009-04-14 19:48:41 +0200192 pipe_lock(pipe);
Jens Axboe5274f052006-03-30 15:15:30 +0200193
Jens Axboe5274f052006-03-30 15:15:30 +0200194 for (;;) {
Ingo Molnar3a326a22006-04-10 15:18:35 +0200195 if (!pipe->readers) {
Jens Axboe5274f052006-03-30 15:15:30 +0200196 send_sig(SIGPIPE, current, 0);
197 if (!ret)
198 ret = -EPIPE;
199 break;
200 }
201
Jens Axboe35f3d142010-05-20 10:43:18 +0200202 if (pipe->nrbufs < pipe->buffers) {
203 int newbuf = (pipe->curbuf + pipe->nrbufs) & (pipe->buffers - 1);
Ingo Molnar3a326a22006-04-10 15:18:35 +0200204 struct pipe_buffer *buf = pipe->bufs + newbuf;
Jens Axboe5274f052006-03-30 15:15:30 +0200205
Jens Axboe912d35f2006-04-26 10:59:21 +0200206 buf->page = spd->pages[page_nr];
207 buf->offset = spd->partial[page_nr].offset;
208 buf->len = spd->partial[page_nr].len;
Jens Axboe497f9622007-06-11 12:00:45 +0200209 buf->private = spd->partial[page_nr].private;
Jens Axboe912d35f2006-04-26 10:59:21 +0200210 buf->ops = spd->ops;
Jens Axboe7afa6fd2006-05-01 20:02:33 +0200211 if (spd->flags & SPLICE_F_GIFT)
212 buf->flags |= PIPE_BUF_FLAG_GIFT;
213
Jens Axboe6f767b02006-04-11 13:53:56 +0200214 pipe->nrbufs++;
Jens Axboe912d35f2006-04-26 10:59:21 +0200215 page_nr++;
216 ret += buf->len;
217
Al Viro6447a3c2013-03-21 11:01:38 -0400218 if (pipe->files)
Jens Axboe6f767b02006-04-11 13:53:56 +0200219 do_wakeup = 1;
Jens Axboe5274f052006-03-30 15:15:30 +0200220
Jens Axboe912d35f2006-04-26 10:59:21 +0200221 if (!--spd->nr_pages)
Jens Axboe5274f052006-03-30 15:15:30 +0200222 break;
Jens Axboe35f3d142010-05-20 10:43:18 +0200223 if (pipe->nrbufs < pipe->buffers)
Jens Axboe5274f052006-03-30 15:15:30 +0200224 continue;
225
226 break;
227 }
228
Jens Axboe912d35f2006-04-26 10:59:21 +0200229 if (spd->flags & SPLICE_F_NONBLOCK) {
Linus Torvalds29e35092006-04-02 12:46:35 -0700230 if (!ret)
231 ret = -EAGAIN;
232 break;
233 }
234
Jens Axboe5274f052006-03-30 15:15:30 +0200235 if (signal_pending(current)) {
236 if (!ret)
237 ret = -ERESTARTSYS;
238 break;
239 }
240
241 if (do_wakeup) {
Jens Axboec0bd1f62006-04-10 09:03:32 +0200242 smp_mb();
Ingo Molnar3a326a22006-04-10 15:18:35 +0200243 if (waitqueue_active(&pipe->wait))
244 wake_up_interruptible_sync(&pipe->wait);
245 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
Jens Axboe5274f052006-03-30 15:15:30 +0200246 do_wakeup = 0;
247 }
248
Ingo Molnar3a326a22006-04-10 15:18:35 +0200249 pipe->waiting_writers++;
250 pipe_wait(pipe);
251 pipe->waiting_writers--;
Jens Axboe5274f052006-03-30 15:15:30 +0200252 }
253
Miklos Szeredi61e0d472009-04-14 19:48:41 +0200254 pipe_unlock(pipe);
Jens Axboe5274f052006-03-30 15:15:30 +0200255
Namhyung Kim825cdcb2011-05-23 19:58:53 +0200256 if (do_wakeup)
257 wakeup_pipe_readers(pipe);
Jens Axboe5274f052006-03-30 15:15:30 +0200258
Jens Axboe00de00b2007-06-15 13:14:22 +0200259 while (page_nr < spd_pages)
Jens Axboebbdfc2f2007-11-06 23:29:47 -0800260 spd->spd_release(spd, page_nr++);
Jens Axboe5274f052006-03-30 15:15:30 +0200261
262 return ret;
263}
Hannes Frederic Sowa2b514572015-05-21 17:00:01 +0200264EXPORT_SYMBOL_GPL(splice_to_pipe);
Jens Axboe5274f052006-03-30 15:15:30 +0200265
Hugh Dickins708e3502011-07-25 17:12:32 -0700266void spd_release_page(struct splice_pipe_desc *spd, unsigned int i)
Jens Axboebbdfc2f2007-11-06 23:29:47 -0800267{
268 page_cache_release(spd->pages[i]);
269}
270
Jens Axboe35f3d142010-05-20 10:43:18 +0200271/*
272 * Check if we need to grow the arrays holding pages and partial page
273 * descriptions.
274 */
Eric Dumazet047fe362012-06-12 15:24:40 +0200275int splice_grow_spd(const struct pipe_inode_info *pipe, struct splice_pipe_desc *spd)
Jens Axboe35f3d142010-05-20 10:43:18 +0200276{
Eric Dumazet047fe362012-06-12 15:24:40 +0200277 unsigned int buffers = ACCESS_ONCE(pipe->buffers);
278
279 spd->nr_pages_max = buffers;
280 if (buffers <= PIPE_DEF_BUFFERS)
Jens Axboe35f3d142010-05-20 10:43:18 +0200281 return 0;
282
Eric Dumazet047fe362012-06-12 15:24:40 +0200283 spd->pages = kmalloc(buffers * sizeof(struct page *), GFP_KERNEL);
284 spd->partial = kmalloc(buffers * sizeof(struct partial_page), GFP_KERNEL);
Jens Axboe35f3d142010-05-20 10:43:18 +0200285
286 if (spd->pages && spd->partial)
287 return 0;
288
289 kfree(spd->pages);
290 kfree(spd->partial);
291 return -ENOMEM;
292}
293
Eric Dumazet047fe362012-06-12 15:24:40 +0200294void splice_shrink_spd(struct splice_pipe_desc *spd)
Jens Axboe35f3d142010-05-20 10:43:18 +0200295{
Eric Dumazet047fe362012-06-12 15:24:40 +0200296 if (spd->nr_pages_max <= PIPE_DEF_BUFFERS)
Jens Axboe35f3d142010-05-20 10:43:18 +0200297 return;
298
299 kfree(spd->pages);
300 kfree(spd->partial);
301}
302
Ingo Molnar3a326a22006-04-10 15:18:35 +0200303static int
Jens Axboecbb7e572006-04-11 14:57:50 +0200304__generic_file_splice_read(struct file *in, loff_t *ppos,
305 struct pipe_inode_info *pipe, size_t len,
306 unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +0200307{
308 struct address_space *mapping = in->f_mapping;
Fengguang Wud8983912007-07-19 01:48:06 -0700309 unsigned int loff, nr_pages, req_pages;
Jens Axboe35f3d142010-05-20 10:43:18 +0200310 struct page *pages[PIPE_DEF_BUFFERS];
311 struct partial_page partial[PIPE_DEF_BUFFERS];
Jens Axboe5274f052006-03-30 15:15:30 +0200312 struct page *page;
Jens Axboe91ad66e2006-04-19 15:55:10 +0200313 pgoff_t index, end_index;
314 loff_t isize;
Jens Axboeeb207962006-04-27 11:05:22 +0200315 int error, page_nr;
Jens Axboe912d35f2006-04-26 10:59:21 +0200316 struct splice_pipe_desc spd = {
317 .pages = pages,
318 .partial = partial,
Eric Dumazet047fe362012-06-12 15:24:40 +0200319 .nr_pages_max = PIPE_DEF_BUFFERS,
Jens Axboe912d35f2006-04-26 10:59:21 +0200320 .flags = flags,
321 .ops = &page_cache_pipe_buf_ops,
Jens Axboebbdfc2f2007-11-06 23:29:47 -0800322 .spd_release = spd_release_page,
Jens Axboe912d35f2006-04-26 10:59:21 +0200323 };
Jens Axboe5274f052006-03-30 15:15:30 +0200324
Jens Axboe35f3d142010-05-20 10:43:18 +0200325 if (splice_grow_spd(pipe, &spd))
326 return -ENOMEM;
327
Jens Axboecbb7e572006-04-11 14:57:50 +0200328 index = *ppos >> PAGE_CACHE_SHIFT;
Jens Axboe912d35f2006-04-26 10:59:21 +0200329 loff = *ppos & ~PAGE_CACHE_MASK;
Fengguang Wud8983912007-07-19 01:48:06 -0700330 req_pages = (len + loff + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
Eric Dumazet047fe362012-06-12 15:24:40 +0200331 nr_pages = min(req_pages, spd.nr_pages_max);
Jens Axboe5274f052006-03-30 15:15:30 +0200332
333 /*
Jens Axboeeb207962006-04-27 11:05:22 +0200334 * Lookup the (hopefully) full range of pages we need.
335 */
Jens Axboe35f3d142010-05-20 10:43:18 +0200336 spd.nr_pages = find_get_pages_contig(mapping, index, nr_pages, spd.pages);
Fengguang Wu431a48202007-07-19 01:48:05 -0700337 index += spd.nr_pages;
Jens Axboeeb207962006-04-27 11:05:22 +0200338
339 /*
340 * If find_get_pages_contig() returned fewer pages than we needed,
Fengguang Wu431a48202007-07-19 01:48:05 -0700341 * readahead/allocate the rest and fill in the holes.
Jens Axboeeb207962006-04-27 11:05:22 +0200342 */
Fengguang Wu431a48202007-07-19 01:48:05 -0700343 if (spd.nr_pages < nr_pages)
Rusty Russellcf914a72007-07-19 01:48:08 -0700344 page_cache_sync_readahead(mapping, &in->f_ra, in,
345 index, req_pages - spd.nr_pages);
Fengguang Wu431a48202007-07-19 01:48:05 -0700346
Jens Axboe932cc6d2007-06-21 13:10:21 +0200347 error = 0;
Jens Axboeeb207962006-04-27 11:05:22 +0200348 while (spd.nr_pages < nr_pages) {
349 /*
350 * Page could be there, find_get_pages_contig() breaks on
351 * the first hole.
352 */
353 page = find_get_page(mapping, index);
354 if (!page) {
Jens Axboee27dedd2006-05-01 19:59:54 +0200355 /*
Jens Axboeeb207962006-04-27 11:05:22 +0200356 * page didn't exist, allocate one.
357 */
358 page = page_cache_alloc_cold(mapping);
359 if (!page)
360 break;
361
362 error = add_to_page_cache_lru(page, mapping, index,
Michal Hockoc62d2552015-11-06 16:28:49 -0800363 mapping_gfp_constraint(mapping, GFP_KERNEL));
Jens Axboeeb207962006-04-27 11:05:22 +0200364 if (unlikely(error)) {
365 page_cache_release(page);
Jens Axboea0548872006-05-03 10:58:22 +0200366 if (error == -EEXIST)
367 continue;
Jens Axboeeb207962006-04-27 11:05:22 +0200368 break;
369 }
370 /*
371 * add_to_page_cache() locks the page, unlock it
372 * to avoid convoluting the logic below even more.
373 */
374 unlock_page(page);
375 }
376
Jens Axboe35f3d142010-05-20 10:43:18 +0200377 spd.pages[spd.nr_pages++] = page;
Jens Axboeeb207962006-04-27 11:05:22 +0200378 index++;
379 }
380
381 /*
382 * Now loop over the map and see if we need to start IO on any
383 * pages, fill in the partial map, etc.
384 */
385 index = *ppos >> PAGE_CACHE_SHIFT;
386 nr_pages = spd.nr_pages;
387 spd.nr_pages = 0;
388 for (page_nr = 0; page_nr < nr_pages; page_nr++) {
Jens Axboe82aa5d62006-04-20 13:05:48 +0200389 unsigned int this_len;
390
391 if (!len)
392 break;
393
394 /*
395 * this_len is the max we'll use from this page
396 */
Andrew Mortonba5f5d92006-04-25 15:33:34 +0200397 this_len = min_t(unsigned long, len, PAGE_CACHE_SIZE - loff);
Jens Axboe35f3d142010-05-20 10:43:18 +0200398 page = spd.pages[page_nr];
Jens Axboe7480a902006-04-11 13:52:47 +0200399
Fengguang Wua08a1662007-07-19 01:48:03 -0700400 if (PageReadahead(page))
Rusty Russellcf914a72007-07-19 01:48:08 -0700401 page_cache_async_readahead(mapping, &in->f_ra, in,
Fengguang Wud8983912007-07-19 01:48:06 -0700402 page, index, req_pages - page_nr);
Fengguang Wua08a1662007-07-19 01:48:03 -0700403
Jens Axboe7480a902006-04-11 13:52:47 +0200404 /*
405 * If the page isn't uptodate, we may need to start io on it
406 */
407 if (!PageUptodate(page)) {
Miklos Szeredi69650312010-08-03 12:48:50 +0200408 lock_page(page);
Jens Axboe7480a902006-04-11 13:52:47 +0200409
410 /*
Miklos Szeredi32502b82008-07-04 09:35:17 +0200411 * Page was truncated, or invalidated by the
412 * filesystem. Redo the find/create, but this time the
413 * page is kept locked, so there's no chance of another
414 * race with truncate/invalidate.
Jens Axboe7480a902006-04-11 13:52:47 +0200415 */
416 if (!page->mapping) {
417 unlock_page(page);
Miklos Szeredi32502b82008-07-04 09:35:17 +0200418 page = find_or_create_page(mapping, index,
419 mapping_gfp_mask(mapping));
420
421 if (!page) {
422 error = -ENOMEM;
423 break;
424 }
Jens Axboe35f3d142010-05-20 10:43:18 +0200425 page_cache_release(spd.pages[page_nr]);
426 spd.pages[page_nr] = page;
Jens Axboe7480a902006-04-11 13:52:47 +0200427 }
428 /*
429 * page was already under io and is now done, great
430 */
431 if (PageUptodate(page)) {
432 unlock_page(page);
433 goto fill_it;
434 }
435
Jens Axboe7480a902006-04-11 13:52:47 +0200436 /*
437 * need to read in the page
438 */
439 error = mapping->a_ops->readpage(in, page);
Jens Axboe7480a902006-04-11 13:52:47 +0200440 if (unlikely(error)) {
Jens Axboeeb207962006-04-27 11:05:22 +0200441 /*
442 * We really should re-lookup the page here,
443 * but it complicates things a lot. Instead
444 * lets just do what we already stored, and
445 * we'll get it the next time we are called.
446 */
Jens Axboe7480a902006-04-11 13:52:47 +0200447 if (error == AOP_TRUNCATED_PAGE)
Jens Axboeeb207962006-04-27 11:05:22 +0200448 error = 0;
449
Jens Axboe7480a902006-04-11 13:52:47 +0200450 break;
451 }
Jens Axboe620a3242007-06-07 09:39:42 +0200452 }
453fill_it:
454 /*
455 * i_size must be checked after PageUptodate.
456 */
457 isize = i_size_read(mapping->host);
458 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
459 if (unlikely(!isize || index > end_index))
460 break;
461
462 /*
463 * if this is the last page, see if we need to shrink
464 * the length and stop
465 */
466 if (end_index == index) {
467 unsigned int plen;
Jens Axboe91ad66e2006-04-19 15:55:10 +0200468
469 /*
Jens Axboe620a3242007-06-07 09:39:42 +0200470 * max good bytes in this page
Jens Axboe91ad66e2006-04-19 15:55:10 +0200471 */
Jens Axboe620a3242007-06-07 09:39:42 +0200472 plen = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
473 if (plen <= loff)
Jens Axboe91ad66e2006-04-19 15:55:10 +0200474 break;
Jens Axboe91ad66e2006-04-19 15:55:10 +0200475
476 /*
Jens Axboe620a3242007-06-07 09:39:42 +0200477 * force quit after adding this page
Jens Axboe91ad66e2006-04-19 15:55:10 +0200478 */
Jens Axboe620a3242007-06-07 09:39:42 +0200479 this_len = min(this_len, plen - loff);
480 len = this_len;
Jens Axboe7480a902006-04-11 13:52:47 +0200481 }
Jens Axboe620a3242007-06-07 09:39:42 +0200482
Jens Axboe35f3d142010-05-20 10:43:18 +0200483 spd.partial[page_nr].offset = loff;
484 spd.partial[page_nr].len = this_len;
Jens Axboe82aa5d62006-04-20 13:05:48 +0200485 len -= this_len;
Jens Axboe91ad66e2006-04-19 15:55:10 +0200486 loff = 0;
Jens Axboeeb207962006-04-27 11:05:22 +0200487 spd.nr_pages++;
488 index++;
Jens Axboe5274f052006-03-30 15:15:30 +0200489 }
490
Jens Axboeeb207962006-04-27 11:05:22 +0200491 /*
Hugh Dickins475ecad2007-06-07 09:36:00 +0200492 * Release any pages at the end, if we quit early. 'page_nr' is how far
Jens Axboeeb207962006-04-27 11:05:22 +0200493 * we got, 'nr_pages' is how many pages are in the map.
494 */
495 while (page_nr < nr_pages)
Jens Axboe35f3d142010-05-20 10:43:18 +0200496 page_cache_release(spd.pages[page_nr++]);
Fengguang Wuf4e6b492007-10-16 01:24:33 -0700497 in->f_ra.prev_pos = (loff_t)index << PAGE_CACHE_SHIFT;
Jens Axboeeb207962006-04-27 11:05:22 +0200498
Jens Axboe912d35f2006-04-26 10:59:21 +0200499 if (spd.nr_pages)
Jens Axboe35f3d142010-05-20 10:43:18 +0200500 error = splice_to_pipe(pipe, &spd);
Jens Axboe5274f052006-03-30 15:15:30 +0200501
Eric Dumazet047fe362012-06-12 15:24:40 +0200502 splice_shrink_spd(&spd);
Jens Axboe7480a902006-04-11 13:52:47 +0200503 return error;
Jens Axboe5274f052006-03-30 15:15:30 +0200504}
505
Jens Axboe83f91352006-04-02 23:05:09 +0200506/**
507 * generic_file_splice_read - splice data from file to a pipe
508 * @in: file to splice from
Jens Axboe932cc6d2007-06-21 13:10:21 +0200509 * @ppos: position in @in
Jens Axboe83f91352006-04-02 23:05:09 +0200510 * @pipe: pipe to splice to
511 * @len: number of bytes to splice
512 * @flags: splice modifier flags
513 *
Jens Axboe932cc6d2007-06-21 13:10:21 +0200514 * Description:
515 * Will read pages from given file and fill them into a pipe. Can be
516 * used as long as the address_space operations for the source implements
517 * a readpage() hook.
518 *
Jens Axboe83f91352006-04-02 23:05:09 +0200519 */
Jens Axboecbb7e572006-04-11 14:57:50 +0200520ssize_t generic_file_splice_read(struct file *in, loff_t *ppos,
521 struct pipe_inode_info *pipe, size_t len,
522 unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +0200523{
Jens Axboed366d3982007-06-01 14:54:11 +0200524 loff_t isize, left;
Jens Axboe8191ecd2008-04-10 08:24:25 +0200525 int ret;
Jens Axboed366d3982007-06-01 14:54:11 +0200526
Boaz Harroshbe64f882015-04-15 16:15:17 -0700527 if (IS_DAX(in->f_mapping->host))
528 return default_file_splice_read(in, ppos, pipe, len, flags);
529
Jens Axboed366d3982007-06-01 14:54:11 +0200530 isize = i_size_read(in->f_mapping->host);
531 if (unlikely(*ppos >= isize))
532 return 0;
533
534 left = isize - *ppos;
535 if (unlikely(left < len))
536 len = left;
Jens Axboe5274f052006-03-30 15:15:30 +0200537
Jens Axboe8191ecd2008-04-10 08:24:25 +0200538 ret = __generic_file_splice_read(in, ppos, pipe, len, flags);
Miklos Szeredi723590e2009-08-15 08:43:22 +0200539 if (ret > 0) {
Jens Axboecbb7e572006-04-11 14:57:50 +0200540 *ppos += ret;
Miklos Szeredi723590e2009-08-15 08:43:22 +0200541 file_accessed(in);
542 }
Jens Axboe5274f052006-03-30 15:15:30 +0200543
544 return ret;
545}
Jens Axboe059a8f32006-04-02 23:06:05 +0200546EXPORT_SYMBOL(generic_file_splice_read);
547
Miklos Szeredi68181732009-05-07 15:37:36 +0200548static const struct pipe_buf_operations default_pipe_buf_ops = {
549 .can_merge = 0,
Miklos Szeredi68181732009-05-07 15:37:36 +0200550 .confirm = generic_pipe_buf_confirm,
551 .release = generic_pipe_buf_release,
552 .steal = generic_pipe_buf_steal,
553 .get = generic_pipe_buf_get,
554};
555
Miklos Szeredi28a625c2014-01-22 19:36:57 +0100556static int generic_pipe_buf_nosteal(struct pipe_inode_info *pipe,
557 struct pipe_buffer *buf)
558{
559 return 1;
560}
561
562/* Pipe buffer operations for a socket and similar. */
563const struct pipe_buf_operations nosteal_pipe_buf_ops = {
564 .can_merge = 0,
Miklos Szeredi28a625c2014-01-22 19:36:57 +0100565 .confirm = generic_pipe_buf_confirm,
566 .release = generic_pipe_buf_release,
567 .steal = generic_pipe_buf_nosteal,
568 .get = generic_pipe_buf_get,
569};
570EXPORT_SYMBOL(nosteal_pipe_buf_ops);
571
Miklos Szeredi68181732009-05-07 15:37:36 +0200572static ssize_t kernel_readv(struct file *file, const struct iovec *vec,
573 unsigned long vlen, loff_t offset)
574{
575 mm_segment_t old_fs;
576 loff_t pos = offset;
577 ssize_t res;
578
579 old_fs = get_fs();
580 set_fs(get_ds());
581 /* The cast to a user pointer is valid due to the set_fs() */
582 res = vfs_readv(file, (const struct iovec __user *)vec, vlen, &pos);
583 set_fs(old_fs);
584
585 return res;
586}
587
Al Viro7bb307e2013-02-23 14:51:48 -0500588ssize_t kernel_write(struct file *file, const char *buf, size_t count,
Miklos Szeredib2858d72009-05-19 11:37:46 +0200589 loff_t pos)
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +0200590{
591 mm_segment_t old_fs;
592 ssize_t res;
593
594 old_fs = get_fs();
595 set_fs(get_ds());
596 /* The cast to a user pointer is valid due to the set_fs() */
Al Viro7bb307e2013-02-23 14:51:48 -0500597 res = vfs_write(file, (__force const char __user *)buf, count, &pos);
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +0200598 set_fs(old_fs);
599
600 return res;
601}
Al Viro7bb307e2013-02-23 14:51:48 -0500602EXPORT_SYMBOL(kernel_write);
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +0200603
Miklos Szeredi68181732009-05-07 15:37:36 +0200604ssize_t default_file_splice_read(struct file *in, loff_t *ppos,
605 struct pipe_inode_info *pipe, size_t len,
606 unsigned int flags)
607{
608 unsigned int nr_pages;
609 unsigned int nr_freed;
610 size_t offset;
Jens Axboe35f3d142010-05-20 10:43:18 +0200611 struct page *pages[PIPE_DEF_BUFFERS];
612 struct partial_page partial[PIPE_DEF_BUFFERS];
613 struct iovec *vec, __vec[PIPE_DEF_BUFFERS];
Miklos Szeredi68181732009-05-07 15:37:36 +0200614 ssize_t res;
615 size_t this_len;
616 int error;
617 int i;
618 struct splice_pipe_desc spd = {
619 .pages = pages,
620 .partial = partial,
Eric Dumazet047fe362012-06-12 15:24:40 +0200621 .nr_pages_max = PIPE_DEF_BUFFERS,
Miklos Szeredi68181732009-05-07 15:37:36 +0200622 .flags = flags,
623 .ops = &default_pipe_buf_ops,
624 .spd_release = spd_release_page,
625 };
626
Jens Axboe35f3d142010-05-20 10:43:18 +0200627 if (splice_grow_spd(pipe, &spd))
628 return -ENOMEM;
629
630 res = -ENOMEM;
631 vec = __vec;
Eric Dumazet047fe362012-06-12 15:24:40 +0200632 if (spd.nr_pages_max > PIPE_DEF_BUFFERS) {
633 vec = kmalloc(spd.nr_pages_max * sizeof(struct iovec), GFP_KERNEL);
Jens Axboe35f3d142010-05-20 10:43:18 +0200634 if (!vec)
635 goto shrink_ret;
636 }
637
Miklos Szeredi68181732009-05-07 15:37:36 +0200638 offset = *ppos & ~PAGE_CACHE_MASK;
639 nr_pages = (len + offset + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
640
Eric Dumazet047fe362012-06-12 15:24:40 +0200641 for (i = 0; i < nr_pages && i < spd.nr_pages_max && len; i++) {
Miklos Szeredi68181732009-05-07 15:37:36 +0200642 struct page *page;
643
Jens Axboe4f231222009-05-13 08:35:35 +0200644 page = alloc_page(GFP_USER);
Miklos Szeredi68181732009-05-07 15:37:36 +0200645 error = -ENOMEM;
646 if (!page)
647 goto err;
648
649 this_len = min_t(size_t, len, PAGE_CACHE_SIZE - offset);
Jens Axboe4f231222009-05-13 08:35:35 +0200650 vec[i].iov_base = (void __user *) page_address(page);
Miklos Szeredi68181732009-05-07 15:37:36 +0200651 vec[i].iov_len = this_len;
Jens Axboe35f3d142010-05-20 10:43:18 +0200652 spd.pages[i] = page;
Miklos Szeredi68181732009-05-07 15:37:36 +0200653 spd.nr_pages++;
654 len -= this_len;
655 offset = 0;
656 }
657
658 res = kernel_readv(in, vec, spd.nr_pages, *ppos);
Andrew Morton77f6bf52009-05-14 09:49:44 +0200659 if (res < 0) {
660 error = res;
Miklos Szeredi68181732009-05-07 15:37:36 +0200661 goto err;
Andrew Morton77f6bf52009-05-14 09:49:44 +0200662 }
Miklos Szeredi68181732009-05-07 15:37:36 +0200663
664 error = 0;
665 if (!res)
666 goto err;
667
668 nr_freed = 0;
669 for (i = 0; i < spd.nr_pages; i++) {
Miklos Szeredi68181732009-05-07 15:37:36 +0200670 this_len = min_t(size_t, vec[i].iov_len, res);
Jens Axboe35f3d142010-05-20 10:43:18 +0200671 spd.partial[i].offset = 0;
672 spd.partial[i].len = this_len;
Miklos Szeredi68181732009-05-07 15:37:36 +0200673 if (!this_len) {
Jens Axboe35f3d142010-05-20 10:43:18 +0200674 __free_page(spd.pages[i]);
675 spd.pages[i] = NULL;
Miklos Szeredi68181732009-05-07 15:37:36 +0200676 nr_freed++;
677 }
678 res -= this_len;
679 }
680 spd.nr_pages -= nr_freed;
681
682 res = splice_to_pipe(pipe, &spd);
683 if (res > 0)
684 *ppos += res;
685
Jens Axboe35f3d142010-05-20 10:43:18 +0200686shrink_ret:
687 if (vec != __vec)
688 kfree(vec);
Eric Dumazet047fe362012-06-12 15:24:40 +0200689 splice_shrink_spd(&spd);
Miklos Szeredi68181732009-05-07 15:37:36 +0200690 return res;
691
692err:
Jens Axboe4f231222009-05-13 08:35:35 +0200693 for (i = 0; i < spd.nr_pages; i++)
Jens Axboe35f3d142010-05-20 10:43:18 +0200694 __free_page(spd.pages[i]);
Jens Axboe4f231222009-05-13 08:35:35 +0200695
Jens Axboe35f3d142010-05-20 10:43:18 +0200696 res = error;
697 goto shrink_ret;
Miklos Szeredi68181732009-05-07 15:37:36 +0200698}
699EXPORT_SYMBOL(default_file_splice_read);
700
Jens Axboe5274f052006-03-30 15:15:30 +0200701/*
Jens Axboe4f6f0bd2006-04-02 23:04:46 +0200702 * Send 'sd->len' bytes to socket from 'sd->file' at position 'sd->pos'
Jens Axboe016b6612006-04-25 15:42:00 +0200703 * using sendpage(). Return the number of bytes sent.
Jens Axboe5274f052006-03-30 15:15:30 +0200704 */
Jens Axboe76ad4d12006-05-03 10:41:33 +0200705static int pipe_to_sendpage(struct pipe_inode_info *pipe,
Jens Axboe5274f052006-03-30 15:15:30 +0200706 struct pipe_buffer *buf, struct splice_desc *sd)
707{
Jens Axboe6a14b902007-06-14 13:08:55 +0200708 struct file *file = sd->u.file;
Jens Axboe5274f052006-03-30 15:15:30 +0200709 loff_t pos = sd->pos;
Michał Mirosława8adbe32010-12-17 08:56:44 +0100710 int more;
Jens Axboe5274f052006-03-30 15:15:30 +0200711
Al Viro72c2d532013-09-22 16:27:52 -0400712 if (!likely(file->f_op->sendpage))
Michał Mirosława8adbe32010-12-17 08:56:44 +0100713 return -EINVAL;
Jens Axboe5274f052006-03-30 15:15:30 +0200714
Eric Dumazet35f9c092012-04-05 03:05:35 +0000715 more = (sd->flags & SPLICE_F_MORE) ? MSG_MORE : 0;
Eric Dumazetae62ca72013-01-06 18:21:49 +0000716
717 if (sd->len < sd->total_len && pipe->nrbufs > 1)
Eric Dumazet35f9c092012-04-05 03:05:35 +0000718 more |= MSG_SENDPAGE_NOTLAST;
Eric Dumazetae62ca72013-01-06 18:21:49 +0000719
Michał Mirosława8adbe32010-12-17 08:56:44 +0100720 return file->f_op->sendpage(file, buf->page, buf->offset,
721 sd->len, &pos, more);
Jens Axboe5274f052006-03-30 15:15:30 +0200722}
723
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200724static void wakeup_pipe_writers(struct pipe_inode_info *pipe)
725{
726 smp_mb();
727 if (waitqueue_active(&pipe->wait))
728 wake_up_interruptible(&pipe->wait);
729 kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
730}
731
732/**
733 * splice_from_pipe_feed - feed available data from a pipe to a file
734 * @pipe: pipe to splice from
735 * @sd: information to @actor
736 * @actor: handler that splices the data
737 *
738 * Description:
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200739 * This function loops over the pipe and calls @actor to do the
740 * actual moving of a single struct pipe_buffer to the desired
741 * destination. It returns when there's no more buffers left in
742 * the pipe or if the requested number of bytes (@sd->total_len)
743 * have been copied. It returns a positive number (one) if the
744 * pipe needs to be filled with more data, zero if the required
745 * number of bytes have been copied and -errno on error.
746 *
747 * This, together with splice_from_pipe_{begin,end,next}, may be
748 * used to implement the functionality of __splice_from_pipe() when
749 * locking is required around copying the pipe buffers to the
750 * destination.
751 */
Al Viro96f9bc82014-04-05 04:35:49 -0400752static int splice_from_pipe_feed(struct pipe_inode_info *pipe, struct splice_desc *sd,
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200753 splice_actor *actor)
754{
755 int ret;
756
757 while (pipe->nrbufs) {
758 struct pipe_buffer *buf = pipe->bufs + pipe->curbuf;
759 const struct pipe_buf_operations *ops = buf->ops;
760
761 sd->len = buf->len;
762 if (sd->len > sd->total_len)
763 sd->len = sd->total_len;
764
Michał Mirosława8adbe32010-12-17 08:56:44 +0100765 ret = buf->ops->confirm(pipe, buf);
766 if (unlikely(ret)) {
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200767 if (ret == -ENODATA)
768 ret = 0;
769 return ret;
770 }
Michał Mirosława8adbe32010-12-17 08:56:44 +0100771
772 ret = actor(pipe, buf, sd);
773 if (ret <= 0)
774 return ret;
775
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200776 buf->offset += ret;
777 buf->len -= ret;
778
779 sd->num_spliced += ret;
780 sd->len -= ret;
781 sd->pos += ret;
782 sd->total_len -= ret;
783
784 if (!buf->len) {
785 buf->ops = NULL;
786 ops->release(pipe, buf);
Jens Axboe35f3d142010-05-20 10:43:18 +0200787 pipe->curbuf = (pipe->curbuf + 1) & (pipe->buffers - 1);
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200788 pipe->nrbufs--;
Al Viro6447a3c2013-03-21 11:01:38 -0400789 if (pipe->files)
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200790 sd->need_wakeup = true;
791 }
792
793 if (!sd->total_len)
794 return 0;
795 }
796
797 return 1;
798}
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200799
800/**
801 * splice_from_pipe_next - wait for some data to splice from
802 * @pipe: pipe to splice from
803 * @sd: information about the splice operation
804 *
805 * Description:
806 * This function will wait for some data and return a positive
807 * value (one) if pipe buffers are available. It will return zero
808 * or -errno if no more data needs to be spliced.
809 */
Al Viro96f9bc82014-04-05 04:35:49 -0400810static int splice_from_pipe_next(struct pipe_inode_info *pipe, struct splice_desc *sd)
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200811{
812 while (!pipe->nrbufs) {
813 if (!pipe->writers)
814 return 0;
815
816 if (!pipe->waiting_writers && sd->num_spliced)
817 return 0;
818
819 if (sd->flags & SPLICE_F_NONBLOCK)
820 return -EAGAIN;
821
822 if (signal_pending(current))
823 return -ERESTARTSYS;
824
825 if (sd->need_wakeup) {
826 wakeup_pipe_writers(pipe);
827 sd->need_wakeup = false;
828 }
829
830 pipe_wait(pipe);
831 }
832
833 return 1;
834}
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200835
836/**
837 * splice_from_pipe_begin - start splicing from pipe
Randy Dunlapb80901b2009-04-16 19:09:55 -0700838 * @sd: information about the splice operation
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200839 *
840 * Description:
841 * This function should be called before a loop containing
842 * splice_from_pipe_next() and splice_from_pipe_feed() to
843 * initialize the necessary fields of @sd.
844 */
Al Viro96f9bc82014-04-05 04:35:49 -0400845static void splice_from_pipe_begin(struct splice_desc *sd)
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200846{
847 sd->num_spliced = 0;
848 sd->need_wakeup = false;
849}
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200850
851/**
852 * splice_from_pipe_end - finish splicing from pipe
853 * @pipe: pipe to splice from
854 * @sd: information about the splice operation
855 *
856 * Description:
857 * This function will wake up pipe writers if necessary. It should
858 * be called after a loop containing splice_from_pipe_next() and
859 * splice_from_pipe_feed().
860 */
Al Viro96f9bc82014-04-05 04:35:49 -0400861static void splice_from_pipe_end(struct pipe_inode_info *pipe, struct splice_desc *sd)
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200862{
863 if (sd->need_wakeup)
864 wakeup_pipe_writers(pipe);
865}
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200866
Jens Axboe932cc6d2007-06-21 13:10:21 +0200867/**
868 * __splice_from_pipe - splice data from a pipe to given actor
869 * @pipe: pipe to splice from
870 * @sd: information to @actor
871 * @actor: handler that splices the data
872 *
873 * Description:
874 * This function does little more than loop over the pipe and call
875 * @actor to do the actual moving of a single struct pipe_buffer to
876 * the desired destination. See pipe_to_file, pipe_to_sendpage, or
877 * pipe_to_user.
878 *
Jens Axboe83f91352006-04-02 23:05:09 +0200879 */
Jens Axboec66ab6f2007-06-12 21:17:17 +0200880ssize_t __splice_from_pipe(struct pipe_inode_info *pipe, struct splice_desc *sd,
881 splice_actor *actor)
Jens Axboe5274f052006-03-30 15:15:30 +0200882{
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200883 int ret;
Jens Axboe5274f052006-03-30 15:15:30 +0200884
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200885 splice_from_pipe_begin(sd);
886 do {
887 ret = splice_from_pipe_next(pipe, sd);
888 if (ret > 0)
889 ret = splice_from_pipe_feed(pipe, sd, actor);
890 } while (ret > 0);
891 splice_from_pipe_end(pipe, sd);
Jens Axboe5274f052006-03-30 15:15:30 +0200892
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200893 return sd->num_spliced ? sd->num_spliced : ret;
Jens Axboe5274f052006-03-30 15:15:30 +0200894}
Mark Fasheh40bee44e2007-03-21 13:11:02 +0100895EXPORT_SYMBOL(__splice_from_pipe);
Jens Axboe5274f052006-03-30 15:15:30 +0200896
Jens Axboe932cc6d2007-06-21 13:10:21 +0200897/**
898 * splice_from_pipe - splice data from a pipe to a file
899 * @pipe: pipe to splice from
900 * @out: file to splice to
901 * @ppos: position in @out
902 * @len: how many bytes to splice
903 * @flags: splice modifier flags
904 * @actor: handler that splices the data
905 *
906 * Description:
Miklos Szeredi29339702009-04-14 19:48:37 +0200907 * See __splice_from_pipe. This function locks the pipe inode,
Jens Axboe932cc6d2007-06-21 13:10:21 +0200908 * otherwise it's identical to __splice_from_pipe().
909 *
910 */
Mark Fasheh6da61802006-10-17 18:43:07 +0200911ssize_t splice_from_pipe(struct pipe_inode_info *pipe, struct file *out,
912 loff_t *ppos, size_t len, unsigned int flags,
913 splice_actor *actor)
914{
915 ssize_t ret;
Jens Axboec66ab6f2007-06-12 21:17:17 +0200916 struct splice_desc sd = {
917 .total_len = len,
918 .flags = flags,
919 .pos = *ppos,
Jens Axboe6a14b902007-06-14 13:08:55 +0200920 .u.file = out,
Jens Axboec66ab6f2007-06-12 21:17:17 +0200921 };
Mark Fasheh6da61802006-10-17 18:43:07 +0200922
Miklos Szeredi61e0d472009-04-14 19:48:41 +0200923 pipe_lock(pipe);
Jens Axboec66ab6f2007-06-12 21:17:17 +0200924 ret = __splice_from_pipe(pipe, &sd, actor);
Miklos Szeredi61e0d472009-04-14 19:48:41 +0200925 pipe_unlock(pipe);
Mark Fasheh6da61802006-10-17 18:43:07 +0200926
927 return ret;
928}
929
930/**
Al Viro8d020762014-04-05 04:27:08 -0400931 * iter_file_splice_write - splice data from a pipe to a file
932 * @pipe: pipe info
933 * @out: file to write to
934 * @ppos: position in @out
935 * @len: number of bytes to splice
936 * @flags: splice modifier flags
937 *
938 * Description:
939 * Will either move or copy pages (determined by @flags options) from
940 * the given pipe inode to the given file.
941 * This one is ->write_iter-based.
942 *
943 */
944ssize_t
945iter_file_splice_write(struct pipe_inode_info *pipe, struct file *out,
946 loff_t *ppos, size_t len, unsigned int flags)
947{
948 struct splice_desc sd = {
949 .total_len = len,
950 .flags = flags,
951 .pos = *ppos,
952 .u.file = out,
953 };
954 int nbufs = pipe->buffers;
955 struct bio_vec *array = kcalloc(nbufs, sizeof(struct bio_vec),
956 GFP_KERNEL);
957 ssize_t ret;
958
959 if (unlikely(!array))
960 return -ENOMEM;
961
962 pipe_lock(pipe);
963
964 splice_from_pipe_begin(&sd);
965 while (sd.total_len) {
966 struct iov_iter from;
Al Viro8d020762014-04-05 04:27:08 -0400967 size_t left;
968 int n, idx;
969
970 ret = splice_from_pipe_next(pipe, &sd);
971 if (ret <= 0)
972 break;
973
974 if (unlikely(nbufs < pipe->buffers)) {
975 kfree(array);
976 nbufs = pipe->buffers;
977 array = kcalloc(nbufs, sizeof(struct bio_vec),
978 GFP_KERNEL);
979 if (!array) {
980 ret = -ENOMEM;
981 break;
982 }
983 }
984
985 /* build the vector */
986 left = sd.total_len;
987 for (n = 0, idx = pipe->curbuf; left && n < pipe->nrbufs; n++, idx++) {
988 struct pipe_buffer *buf = pipe->bufs + idx;
989 size_t this_len = buf->len;
990
991 if (this_len > left)
992 this_len = left;
993
994 if (idx == pipe->buffers - 1)
995 idx = -1;
996
997 ret = buf->ops->confirm(pipe, buf);
998 if (unlikely(ret)) {
999 if (ret == -ENODATA)
1000 ret = 0;
1001 goto done;
1002 }
1003
1004 array[n].bv_page = buf->page;
1005 array[n].bv_len = this_len;
1006 array[n].bv_offset = buf->offset;
1007 left -= this_len;
1008 }
1009
Al Viro05afcb72015-01-23 01:08:07 -05001010 iov_iter_bvec(&from, ITER_BVEC | WRITE, array, n,
1011 sd.total_len - left);
Christoph Hellwigdbe4e192015-01-25 21:11:59 +01001012 ret = vfs_iter_write(out, &from, &sd.pos);
Al Viro8d020762014-04-05 04:27:08 -04001013 if (ret <= 0)
1014 break;
1015
1016 sd.num_spliced += ret;
1017 sd.total_len -= ret;
Christoph Hellwigdbe4e192015-01-25 21:11:59 +01001018 *ppos = sd.pos;
Al Viro8d020762014-04-05 04:27:08 -04001019
1020 /* dismiss the fully eaten buffers, adjust the partial one */
1021 while (ret) {
1022 struct pipe_buffer *buf = pipe->bufs + pipe->curbuf;
1023 if (ret >= buf->len) {
1024 const struct pipe_buf_operations *ops = buf->ops;
1025 ret -= buf->len;
1026 buf->len = 0;
1027 buf->ops = NULL;
1028 ops->release(pipe, buf);
1029 pipe->curbuf = (pipe->curbuf + 1) & (pipe->buffers - 1);
1030 pipe->nrbufs--;
1031 if (pipe->files)
1032 sd.need_wakeup = true;
1033 } else {
1034 buf->offset += ret;
1035 buf->len -= ret;
1036 ret = 0;
1037 }
1038 }
1039 }
1040done:
1041 kfree(array);
1042 splice_from_pipe_end(pipe, &sd);
1043
1044 pipe_unlock(pipe);
1045
1046 if (sd.num_spliced)
1047 ret = sd.num_spliced;
1048
1049 return ret;
1050}
1051
1052EXPORT_SYMBOL(iter_file_splice_write);
1053
Miklos Szeredib2858d72009-05-19 11:37:46 +02001054static int write_pipe_buf(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
1055 struct splice_desc *sd)
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +02001056{
Miklos Szeredib2858d72009-05-19 11:37:46 +02001057 int ret;
1058 void *data;
Al Viro06ae43f2013-03-20 13:19:30 -04001059 loff_t tmp = sd->pos;
Miklos Szeredib2858d72009-05-19 11:37:46 +02001060
Al Virofbb32752014-02-02 21:09:54 -05001061 data = kmap(buf->page);
Al Viro06ae43f2013-03-20 13:19:30 -04001062 ret = __kernel_write(sd->u.file, data + buf->offset, sd->len, &tmp);
Al Virofbb32752014-02-02 21:09:54 -05001063 kunmap(buf->page);
Miklos Szeredib2858d72009-05-19 11:37:46 +02001064
1065 return ret;
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +02001066}
1067
1068static ssize_t default_file_splice_write(struct pipe_inode_info *pipe,
1069 struct file *out, loff_t *ppos,
1070 size_t len, unsigned int flags)
1071{
Miklos Szeredib2858d72009-05-19 11:37:46 +02001072 ssize_t ret;
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +02001073
Miklos Szeredib2858d72009-05-19 11:37:46 +02001074 ret = splice_from_pipe(pipe, out, ppos, len, flags, write_pipe_buf);
1075 if (ret > 0)
1076 *ppos += ret;
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +02001077
Miklos Szeredib2858d72009-05-19 11:37:46 +02001078 return ret;
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +02001079}
1080
Jens Axboe83f91352006-04-02 23:05:09 +02001081/**
1082 * generic_splice_sendpage - splice data from a pipe to a socket
Jens Axboe932cc6d2007-06-21 13:10:21 +02001083 * @pipe: pipe to splice from
Jens Axboe83f91352006-04-02 23:05:09 +02001084 * @out: socket to write to
Jens Axboe932cc6d2007-06-21 13:10:21 +02001085 * @ppos: position in @out
Jens Axboe83f91352006-04-02 23:05:09 +02001086 * @len: number of bytes to splice
1087 * @flags: splice modifier flags
1088 *
Jens Axboe932cc6d2007-06-21 13:10:21 +02001089 * Description:
1090 * Will send @len bytes from the pipe to a network socket. No data copying
1091 * is involved.
Jens Axboe83f91352006-04-02 23:05:09 +02001092 *
1093 */
Ingo Molnar3a326a22006-04-10 15:18:35 +02001094ssize_t generic_splice_sendpage(struct pipe_inode_info *pipe, struct file *out,
Jens Axboecbb7e572006-04-11 14:57:50 +02001095 loff_t *ppos, size_t len, unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +02001096{
Jens Axboe00522fb2006-04-26 14:39:29 +02001097 return splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_sendpage);
Jens Axboe5274f052006-03-30 15:15:30 +02001098}
1099
Jens Axboe059a8f32006-04-02 23:06:05 +02001100EXPORT_SYMBOL(generic_splice_sendpage);
Jeff Garzika0f06782006-03-30 23:06:13 -05001101
Jens Axboe83f91352006-04-02 23:05:09 +02001102/*
1103 * Attempt to initiate a splice from pipe to file.
1104 */
Ingo Molnar3a326a22006-04-10 15:18:35 +02001105static long do_splice_from(struct pipe_inode_info *pipe, struct file *out,
Jens Axboecbb7e572006-04-11 14:57:50 +02001106 loff_t *ppos, size_t len, unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +02001107{
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +02001108 ssize_t (*splice_write)(struct pipe_inode_info *, struct file *,
1109 loff_t *, size_t, unsigned int);
Jens Axboe5274f052006-03-30 15:15:30 +02001110
Al Viro72c2d532013-09-22 16:27:52 -04001111 if (out->f_op->splice_write)
Changli Gaocc56f7d2009-11-04 09:09:52 +01001112 splice_write = out->f_op->splice_write;
1113 else
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +02001114 splice_write = default_file_splice_write;
1115
Al Viro500368f2013-05-23 20:07:11 -04001116 return splice_write(pipe, out, ppos, len, flags);
Jens Axboe5274f052006-03-30 15:15:30 +02001117}
1118
Jens Axboe83f91352006-04-02 23:05:09 +02001119/*
1120 * Attempt to initiate a splice from a file to a pipe.
1121 */
Jens Axboecbb7e572006-04-11 14:57:50 +02001122static long do_splice_to(struct file *in, loff_t *ppos,
1123 struct pipe_inode_info *pipe, size_t len,
1124 unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +02001125{
Miklos Szeredi68181732009-05-07 15:37:36 +02001126 ssize_t (*splice_read)(struct file *, loff_t *,
1127 struct pipe_inode_info *, size_t, unsigned int);
Jens Axboe5274f052006-03-30 15:15:30 +02001128 int ret;
1129
Jens Axboe49570e92006-04-11 13:56:09 +02001130 if (unlikely(!(in->f_mode & FMODE_READ)))
Jens Axboe5274f052006-03-30 15:15:30 +02001131 return -EBADF;
1132
Jens Axboecbb7e572006-04-11 14:57:50 +02001133 ret = rw_verify_area(READ, in, ppos, len);
Jens Axboe5274f052006-03-30 15:15:30 +02001134 if (unlikely(ret < 0))
1135 return ret;
1136
Al Viro72c2d532013-09-22 16:27:52 -04001137 if (in->f_op->splice_read)
Changli Gaocc56f7d2009-11-04 09:09:52 +01001138 splice_read = in->f_op->splice_read;
1139 else
Miklos Szeredi68181732009-05-07 15:37:36 +02001140 splice_read = default_file_splice_read;
1141
1142 return splice_read(in, ppos, pipe, len, flags);
Jens Axboe5274f052006-03-30 15:15:30 +02001143}
1144
Jens Axboe932cc6d2007-06-21 13:10:21 +02001145/**
1146 * splice_direct_to_actor - splices data directly between two non-pipes
1147 * @in: file to splice from
1148 * @sd: actor information on where to splice to
1149 * @actor: handles the data splicing
1150 *
1151 * Description:
1152 * This is a special case helper to splice directly between two
1153 * points, without requiring an explicit pipe. Internally an allocated
Randy Dunlap79685b82007-07-27 08:08:51 +02001154 * pipe is cached in the process, and reused during the lifetime of
Jens Axboe932cc6d2007-06-21 13:10:21 +02001155 * that process.
1156 *
Jens Axboec66ab6f2007-06-12 21:17:17 +02001157 */
1158ssize_t splice_direct_to_actor(struct file *in, struct splice_desc *sd,
1159 splice_direct_actor *actor)
Jens Axboeb92ce552006-04-11 13:52:07 +02001160{
1161 struct pipe_inode_info *pipe;
1162 long ret, bytes;
1163 umode_t i_mode;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001164 size_t len;
Christophe Leroy0ff28d92015-05-06 17:26:47 +02001165 int i, flags, more;
Jens Axboeb92ce552006-04-11 13:52:07 +02001166
1167 /*
1168 * We require the input being a regular file, as we don't want to
1169 * randomly drop data for eg socket -> socket splicing. Use the
1170 * piped splicing for that!
1171 */
Al Viro496ad9a2013-01-23 17:07:38 -05001172 i_mode = file_inode(in)->i_mode;
Jens Axboeb92ce552006-04-11 13:52:07 +02001173 if (unlikely(!S_ISREG(i_mode) && !S_ISBLK(i_mode)))
1174 return -EINVAL;
1175
1176 /*
1177 * neither in nor out is a pipe, setup an internal pipe attached to
1178 * 'out' and transfer the wanted data from 'in' to 'out' through that
1179 */
1180 pipe = current->splice_pipe;
Jens Axboe49570e92006-04-11 13:56:09 +02001181 if (unlikely(!pipe)) {
Al Viro7bee1302013-03-21 11:04:15 -04001182 pipe = alloc_pipe_info();
Jens Axboeb92ce552006-04-11 13:52:07 +02001183 if (!pipe)
1184 return -ENOMEM;
1185
1186 /*
1187 * We don't have an immediate reader, but we'll read the stuff
Jens Axboe00522fb2006-04-26 14:39:29 +02001188 * out of the pipe right after the splice_to_pipe(). So set
Jens Axboeb92ce552006-04-11 13:52:07 +02001189 * PIPE_READERS appropriately.
1190 */
1191 pipe->readers = 1;
1192
1193 current->splice_pipe = pipe;
1194 }
1195
1196 /*
Ingo Molnar73d62d82006-04-11 13:57:21 +02001197 * Do the splice.
Jens Axboeb92ce552006-04-11 13:52:07 +02001198 */
1199 ret = 0;
1200 bytes = 0;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001201 len = sd->total_len;
1202 flags = sd->flags;
1203
1204 /*
1205 * Don't block on output, we have to drain the direct pipe.
1206 */
1207 sd->flags &= ~SPLICE_F_NONBLOCK;
Christophe Leroy0ff28d92015-05-06 17:26:47 +02001208 more = sd->flags & SPLICE_F_MORE;
Jens Axboeb92ce552006-04-11 13:52:07 +02001209
1210 while (len) {
Jens Axboe51a92c02007-07-13 14:11:43 +02001211 size_t read_len;
Tom Zanussia82c53a2008-05-09 13:28:36 +02001212 loff_t pos = sd->pos, prev_pos = pos;
Jens Axboeb92ce552006-04-11 13:52:07 +02001213
Jens Axboebcd4f3a2007-07-16 14:41:49 +02001214 ret = do_splice_to(in, &pos, pipe, len, flags);
Jens Axboe51a92c02007-07-13 14:11:43 +02001215 if (unlikely(ret <= 0))
Jens Axboeb92ce552006-04-11 13:52:07 +02001216 goto out_release;
1217
1218 read_len = ret;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001219 sd->total_len = read_len;
Jens Axboeb92ce552006-04-11 13:52:07 +02001220
1221 /*
Christophe Leroy0ff28d92015-05-06 17:26:47 +02001222 * If more data is pending, set SPLICE_F_MORE
1223 * If this is the last data and SPLICE_F_MORE was not set
1224 * initially, clears it.
1225 */
1226 if (read_len < len)
1227 sd->flags |= SPLICE_F_MORE;
1228 else if (!more)
1229 sd->flags &= ~SPLICE_F_MORE;
1230 /*
Jens Axboeb92ce552006-04-11 13:52:07 +02001231 * NOTE: nonblocking mode only applies to the input. We
1232 * must not do the output in nonblocking mode as then we
1233 * could get stuck data in the internal pipe:
1234 */
Jens Axboec66ab6f2007-06-12 21:17:17 +02001235 ret = actor(pipe, sd);
Tom Zanussia82c53a2008-05-09 13:28:36 +02001236 if (unlikely(ret <= 0)) {
1237 sd->pos = prev_pos;
Jens Axboeb92ce552006-04-11 13:52:07 +02001238 goto out_release;
Tom Zanussia82c53a2008-05-09 13:28:36 +02001239 }
Jens Axboeb92ce552006-04-11 13:52:07 +02001240
1241 bytes += ret;
1242 len -= ret;
Jens Axboebcd4f3a2007-07-16 14:41:49 +02001243 sd->pos = pos;
Jens Axboeb92ce552006-04-11 13:52:07 +02001244
Tom Zanussia82c53a2008-05-09 13:28:36 +02001245 if (ret < read_len) {
1246 sd->pos = prev_pos + ret;
Jens Axboe51a92c02007-07-13 14:11:43 +02001247 goto out_release;
Tom Zanussia82c53a2008-05-09 13:28:36 +02001248 }
Jens Axboeb92ce552006-04-11 13:52:07 +02001249 }
1250
Jens Axboe9e971982008-01-29 21:05:57 +01001251done:
Jens Axboeb92ce552006-04-11 13:52:07 +02001252 pipe->nrbufs = pipe->curbuf = 0;
Jens Axboe80848702008-01-30 12:24:48 +01001253 file_accessed(in);
Jens Axboeb92ce552006-04-11 13:52:07 +02001254 return bytes;
1255
1256out_release:
1257 /*
1258 * If we did an incomplete transfer we must release
1259 * the pipe buffers in question:
1260 */
Jens Axboe35f3d142010-05-20 10:43:18 +02001261 for (i = 0; i < pipe->buffers; i++) {
Jens Axboeb92ce552006-04-11 13:52:07 +02001262 struct pipe_buffer *buf = pipe->bufs + i;
1263
1264 if (buf->ops) {
1265 buf->ops->release(pipe, buf);
1266 buf->ops = NULL;
1267 }
1268 }
Jens Axboeb92ce552006-04-11 13:52:07 +02001269
Jens Axboe9e971982008-01-29 21:05:57 +01001270 if (!bytes)
1271 bytes = ret;
Jens Axboeb92ce552006-04-11 13:52:07 +02001272
Jens Axboe9e971982008-01-29 21:05:57 +01001273 goto done;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001274}
1275EXPORT_SYMBOL(splice_direct_to_actor);
1276
1277static int direct_splice_actor(struct pipe_inode_info *pipe,
1278 struct splice_desc *sd)
1279{
Jens Axboe6a14b902007-06-14 13:08:55 +02001280 struct file *file = sd->u.file;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001281
Al Viro7995bd22013-06-20 18:58:36 +04001282 return do_splice_from(pipe, file, sd->opos, sd->total_len,
Changli Gao2cb4b052010-06-29 13:09:18 +02001283 sd->flags);
Jens Axboec66ab6f2007-06-12 21:17:17 +02001284}
1285
Jens Axboe932cc6d2007-06-21 13:10:21 +02001286/**
1287 * do_splice_direct - splices data directly between two files
1288 * @in: file to splice from
1289 * @ppos: input file offset
1290 * @out: file to splice to
Randy Dunlapacdb37c2013-06-22 19:44:08 -07001291 * @opos: output file offset
Jens Axboe932cc6d2007-06-21 13:10:21 +02001292 * @len: number of bytes to splice
1293 * @flags: splice modifier flags
1294 *
1295 * Description:
1296 * For use by do_sendfile(). splice can easily emulate sendfile, but
1297 * doing it in the application would incur an extra system call
1298 * (splice in + splice out, as compared to just sendfile()). So this helper
1299 * can splice directly through a process-private pipe.
1300 *
1301 */
Jens Axboec66ab6f2007-06-12 21:17:17 +02001302long do_splice_direct(struct file *in, loff_t *ppos, struct file *out,
Al Viro7995bd22013-06-20 18:58:36 +04001303 loff_t *opos, size_t len, unsigned int flags)
Jens Axboec66ab6f2007-06-12 21:17:17 +02001304{
1305 struct splice_desc sd = {
1306 .len = len,
1307 .total_len = len,
1308 .flags = flags,
1309 .pos = *ppos,
Jens Axboe6a14b902007-06-14 13:08:55 +02001310 .u.file = out,
Al Viro7995bd22013-06-20 18:58:36 +04001311 .opos = opos,
Jens Axboec66ab6f2007-06-12 21:17:17 +02001312 };
Jens Axboe51a92c02007-07-13 14:11:43 +02001313 long ret;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001314
Al Viro18c67cb2013-06-19 15:41:54 +04001315 if (unlikely(!(out->f_mode & FMODE_WRITE)))
1316 return -EBADF;
1317
1318 if (unlikely(out->f_flags & O_APPEND))
1319 return -EINVAL;
1320
1321 ret = rw_verify_area(WRITE, out, opos, len);
1322 if (unlikely(ret < 0))
1323 return ret;
1324
Jens Axboec66ab6f2007-06-12 21:17:17 +02001325 ret = splice_direct_to_actor(in, &sd, direct_splice_actor);
Jens Axboe51a92c02007-07-13 14:11:43 +02001326 if (ret > 0)
Tom Zanussia82c53a2008-05-09 13:28:36 +02001327 *ppos = sd.pos;
Jens Axboe51a92c02007-07-13 14:11:43 +02001328
Jens Axboec66ab6f2007-06-12 21:17:17 +02001329 return ret;
Jens Axboeb92ce552006-04-11 13:52:07 +02001330}
Miklos Szeredi1c118592014-10-24 00:14:35 +02001331EXPORT_SYMBOL(do_splice_direct);
Jens Axboeb92ce552006-04-11 13:52:07 +02001332
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001333static int splice_pipe_to_pipe(struct pipe_inode_info *ipipe,
1334 struct pipe_inode_info *opipe,
1335 size_t len, unsigned int flags);
Jens Axboeddac0d32006-11-04 12:49:32 +01001336
1337/*
Jens Axboe83f91352006-04-02 23:05:09 +02001338 * Determine where to splice to/from.
1339 */
Ingo Molnar529565d2006-04-10 15:18:58 +02001340static long do_splice(struct file *in, loff_t __user *off_in,
1341 struct file *out, loff_t __user *off_out,
1342 size_t len, unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +02001343{
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001344 struct pipe_inode_info *ipipe;
1345 struct pipe_inode_info *opipe;
Al Viro7995bd22013-06-20 18:58:36 +04001346 loff_t offset;
Jens Axboea4514eb2006-04-19 15:57:05 +02001347 long ret;
Jens Axboe5274f052006-03-30 15:15:30 +02001348
Linus Torvalds71993e62010-11-28 13:56:09 -08001349 ipipe = get_pipe_info(in);
1350 opipe = get_pipe_info(out);
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001351
1352 if (ipipe && opipe) {
1353 if (off_in || off_out)
1354 return -ESPIPE;
1355
1356 if (!(in->f_mode & FMODE_READ))
1357 return -EBADF;
1358
1359 if (!(out->f_mode & FMODE_WRITE))
1360 return -EBADF;
1361
1362 /* Splicing to self would be fun, but... */
1363 if (ipipe == opipe)
1364 return -EINVAL;
1365
1366 return splice_pipe_to_pipe(ipipe, opipe, len, flags);
1367 }
1368
1369 if (ipipe) {
Ingo Molnar529565d2006-04-10 15:18:58 +02001370 if (off_in)
1371 return -ESPIPE;
Jens Axboeb92ce552006-04-11 13:52:07 +02001372 if (off_out) {
Changli Gao19c9a492010-06-29 13:10:36 +02001373 if (!(out->f_mode & FMODE_PWRITE))
Jens Axboeb92ce552006-04-11 13:52:07 +02001374 return -EINVAL;
Jens Axboecbb7e572006-04-11 14:57:50 +02001375 if (copy_from_user(&offset, off_out, sizeof(loff_t)))
Jens Axboeb92ce552006-04-11 13:52:07 +02001376 return -EFAULT;
Al Viro7995bd22013-06-20 18:58:36 +04001377 } else {
1378 offset = out->f_pos;
1379 }
Ingo Molnar529565d2006-04-10 15:18:58 +02001380
Al Viro18c67cb2013-06-19 15:41:54 +04001381 if (unlikely(!(out->f_mode & FMODE_WRITE)))
1382 return -EBADF;
1383
1384 if (unlikely(out->f_flags & O_APPEND))
1385 return -EINVAL;
1386
1387 ret = rw_verify_area(WRITE, out, &offset, len);
1388 if (unlikely(ret < 0))
1389 return ret;
1390
Al Viro500368f2013-05-23 20:07:11 -04001391 file_start_write(out);
Al Viro7995bd22013-06-20 18:58:36 +04001392 ret = do_splice_from(ipipe, out, &offset, len, flags);
Al Viro500368f2013-05-23 20:07:11 -04001393 file_end_write(out);
Jens Axboea4514eb2006-04-19 15:57:05 +02001394
Al Viro7995bd22013-06-20 18:58:36 +04001395 if (!off_out)
1396 out->f_pos = offset;
1397 else if (copy_to_user(off_out, &offset, sizeof(loff_t)))
Jens Axboea4514eb2006-04-19 15:57:05 +02001398 ret = -EFAULT;
1399
1400 return ret;
Ingo Molnar529565d2006-04-10 15:18:58 +02001401 }
Jens Axboe5274f052006-03-30 15:15:30 +02001402
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001403 if (opipe) {
Ingo Molnar529565d2006-04-10 15:18:58 +02001404 if (off_out)
1405 return -ESPIPE;
Jens Axboeb92ce552006-04-11 13:52:07 +02001406 if (off_in) {
Changli Gao19c9a492010-06-29 13:10:36 +02001407 if (!(in->f_mode & FMODE_PREAD))
Jens Axboeb92ce552006-04-11 13:52:07 +02001408 return -EINVAL;
Jens Axboecbb7e572006-04-11 14:57:50 +02001409 if (copy_from_user(&offset, off_in, sizeof(loff_t)))
Jens Axboeb92ce552006-04-11 13:52:07 +02001410 return -EFAULT;
Al Viro7995bd22013-06-20 18:58:36 +04001411 } else {
1412 offset = in->f_pos;
1413 }
Ingo Molnar529565d2006-04-10 15:18:58 +02001414
Al Viro7995bd22013-06-20 18:58:36 +04001415 ret = do_splice_to(in, &offset, opipe, len, flags);
Jens Axboea4514eb2006-04-19 15:57:05 +02001416
Al Viro7995bd22013-06-20 18:58:36 +04001417 if (!off_in)
1418 in->f_pos = offset;
1419 else if (copy_to_user(off_in, &offset, sizeof(loff_t)))
Jens Axboea4514eb2006-04-19 15:57:05 +02001420 ret = -EFAULT;
1421
1422 return ret;
Ingo Molnar529565d2006-04-10 15:18:58 +02001423 }
Jens Axboe5274f052006-03-30 15:15:30 +02001424
1425 return -EINVAL;
1426}
1427
Jens Axboe912d35f2006-04-26 10:59:21 +02001428/*
1429 * Map an iov into an array of pages and offset/length tupples. With the
1430 * partial_page structure, we can map several non-contiguous ranges into
1431 * our ones pages[] map instead of splitting that operation into pieces.
1432 * Could easily be exported as a generic helper for other users, in which
1433 * case one would probably want to add a 'max_nr_pages' parameter as well.
1434 */
1435static int get_iovec_page_array(const struct iovec __user *iov,
1436 unsigned int nr_vecs, struct page **pages,
Eric Dumazetbd1a68b2012-04-04 11:08:51 +02001437 struct partial_page *partial, bool aligned,
Jens Axboe35f3d142010-05-20 10:43:18 +02001438 unsigned int pipe_buffers)
Jens Axboe912d35f2006-04-26 10:59:21 +02001439{
1440 int buffers = 0, error = 0;
1441
Jens Axboe912d35f2006-04-26 10:59:21 +02001442 while (nr_vecs) {
1443 unsigned long off, npages;
Linus Torvalds75723952007-10-01 13:17:28 -07001444 struct iovec entry;
Jens Axboe912d35f2006-04-26 10:59:21 +02001445 void __user *base;
1446 size_t len;
1447 int i;
1448
Linus Torvalds75723952007-10-01 13:17:28 -07001449 error = -EFAULT;
Nick Pigginbc40d732008-07-25 19:45:26 -07001450 if (copy_from_user(&entry, iov, sizeof(entry)))
Jens Axboe912d35f2006-04-26 10:59:21 +02001451 break;
Linus Torvalds75723952007-10-01 13:17:28 -07001452
1453 base = entry.iov_base;
1454 len = entry.iov_len;
Jens Axboe912d35f2006-04-26 10:59:21 +02001455
1456 /*
1457 * Sanity check this iovec. 0 read succeeds.
1458 */
Linus Torvalds75723952007-10-01 13:17:28 -07001459 error = 0;
Jens Axboe912d35f2006-04-26 10:59:21 +02001460 if (unlikely(!len))
1461 break;
1462 error = -EFAULT;
Bastian Blank712a30e2008-02-10 16:47:57 +02001463 if (!access_ok(VERIFY_READ, base, len))
Jens Axboe912d35f2006-04-26 10:59:21 +02001464 break;
1465
1466 /*
1467 * Get this base offset and number of pages, then map
1468 * in the user pages.
1469 */
1470 off = (unsigned long) base & ~PAGE_MASK;
Jens Axboe7afa6fd2006-05-01 20:02:33 +02001471
1472 /*
1473 * If asked for alignment, the offset must be zero and the
1474 * length a multiple of the PAGE_SIZE.
1475 */
1476 error = -EINVAL;
1477 if (aligned && (off || len & ~PAGE_MASK))
1478 break;
1479
Jens Axboe912d35f2006-04-26 10:59:21 +02001480 npages = (off + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
Jens Axboe35f3d142010-05-20 10:43:18 +02001481 if (npages > pipe_buffers - buffers)
1482 npages = pipe_buffers - buffers;
Jens Axboe912d35f2006-04-26 10:59:21 +02001483
Nick Pigginbc40d732008-07-25 19:45:26 -07001484 error = get_user_pages_fast((unsigned long)base, npages,
1485 0, &pages[buffers]);
Jens Axboe912d35f2006-04-26 10:59:21 +02001486
1487 if (unlikely(error <= 0))
1488 break;
1489
1490 /*
1491 * Fill this contiguous range into the partial page map.
1492 */
1493 for (i = 0; i < error; i++) {
Jens Axboe75914892006-05-02 12:57:18 +02001494 const int plen = min_t(size_t, len, PAGE_SIZE - off);
Jens Axboe912d35f2006-04-26 10:59:21 +02001495
1496 partial[buffers].offset = off;
1497 partial[buffers].len = plen;
1498
1499 off = 0;
1500 len -= plen;
1501 buffers++;
1502 }
1503
1504 /*
1505 * We didn't complete this iov, stop here since it probably
1506 * means we have to move some of this into a pipe to
1507 * be able to continue.
1508 */
1509 if (len)
1510 break;
1511
1512 /*
1513 * Don't continue if we mapped fewer pages than we asked for,
1514 * or if we mapped the max number of pages that we have
1515 * room for.
1516 */
Jens Axboe35f3d142010-05-20 10:43:18 +02001517 if (error < npages || buffers == pipe_buffers)
Jens Axboe912d35f2006-04-26 10:59:21 +02001518 break;
1519
1520 nr_vecs--;
1521 iov++;
1522 }
1523
Jens Axboe912d35f2006-04-26 10:59:21 +02001524 if (buffers)
1525 return buffers;
1526
1527 return error;
1528}
1529
Jens Axboe6a14b902007-06-14 13:08:55 +02001530static int pipe_to_user(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
1531 struct splice_desc *sd)
1532{
Al Viro6130f532014-02-03 18:19:51 -05001533 int n = copy_page_to_iter(buf->page, buf->offset, sd->len, sd->u.data);
1534 return n == sd->len ? n : -EFAULT;
Jens Axboe6a14b902007-06-14 13:08:55 +02001535}
1536
1537/*
1538 * For lack of a better implementation, implement vmsplice() to userspace
1539 * as a simple copy of the pipes pages to the user iov.
1540 */
Al Viro6130f532014-02-03 18:19:51 -05001541static long vmsplice_to_user(struct file *file, const struct iovec __user *uiov,
Jens Axboe6a14b902007-06-14 13:08:55 +02001542 unsigned long nr_segs, unsigned int flags)
1543{
1544 struct pipe_inode_info *pipe;
1545 struct splice_desc sd;
Jens Axboe6a14b902007-06-14 13:08:55 +02001546 long ret;
Al Viro6130f532014-02-03 18:19:51 -05001547 struct iovec iovstack[UIO_FASTIOV];
1548 struct iovec *iov = iovstack;
1549 struct iov_iter iter;
Jens Axboe6a14b902007-06-14 13:08:55 +02001550
Linus Torvalds71993e62010-11-28 13:56:09 -08001551 pipe = get_pipe_info(file);
Jens Axboe6a14b902007-06-14 13:08:55 +02001552 if (!pipe)
1553 return -EBADF;
1554
Al Viro345995f2015-03-21 19:17:55 -04001555 ret = import_iovec(READ, uiov, nr_segs,
1556 ARRAY_SIZE(iovstack), &iov, &iter);
1557 if (ret < 0)
1558 return ret;
Al Viro6130f532014-02-03 18:19:51 -05001559
Al Viro345995f2015-03-21 19:17:55 -04001560 sd.total_len = iov_iter_count(&iter);
Al Viro6130f532014-02-03 18:19:51 -05001561 sd.len = 0;
Al Viro6130f532014-02-03 18:19:51 -05001562 sd.flags = flags;
1563 sd.u.data = &iter;
1564 sd.pos = 0;
1565
Al Viro345995f2015-03-21 19:17:55 -04001566 if (sd.total_len) {
1567 pipe_lock(pipe);
1568 ret = __splice_from_pipe(pipe, &sd, pipe_to_user);
1569 pipe_unlock(pipe);
1570 }
Jens Axboe6a14b902007-06-14 13:08:55 +02001571
Al Viro345995f2015-03-21 19:17:55 -04001572 kfree(iov);
Jens Axboe6a14b902007-06-14 13:08:55 +02001573 return ret;
1574}
1575
Jens Axboe912d35f2006-04-26 10:59:21 +02001576/*
1577 * vmsplice splices a user address range into a pipe. It can be thought of
1578 * as splice-from-memory, where the regular splice is splice-from-file (or
1579 * to file). In both cases the output is a pipe, naturally.
Jens Axboe912d35f2006-04-26 10:59:21 +02001580 */
Jens Axboe6a14b902007-06-14 13:08:55 +02001581static long vmsplice_to_pipe(struct file *file, const struct iovec __user *iov,
1582 unsigned long nr_segs, unsigned int flags)
Jens Axboe912d35f2006-04-26 10:59:21 +02001583{
Jens Axboeddac0d32006-11-04 12:49:32 +01001584 struct pipe_inode_info *pipe;
Jens Axboe35f3d142010-05-20 10:43:18 +02001585 struct page *pages[PIPE_DEF_BUFFERS];
1586 struct partial_page partial[PIPE_DEF_BUFFERS];
Jens Axboe912d35f2006-04-26 10:59:21 +02001587 struct splice_pipe_desc spd = {
1588 .pages = pages,
1589 .partial = partial,
Eric Dumazet047fe362012-06-12 15:24:40 +02001590 .nr_pages_max = PIPE_DEF_BUFFERS,
Jens Axboe912d35f2006-04-26 10:59:21 +02001591 .flags = flags,
1592 .ops = &user_page_pipe_buf_ops,
Jens Axboebbdfc2f2007-11-06 23:29:47 -08001593 .spd_release = spd_release_page,
Jens Axboe912d35f2006-04-26 10:59:21 +02001594 };
Jens Axboe35f3d142010-05-20 10:43:18 +02001595 long ret;
Jens Axboe912d35f2006-04-26 10:59:21 +02001596
Linus Torvalds71993e62010-11-28 13:56:09 -08001597 pipe = get_pipe_info(file);
Jens Axboeddac0d32006-11-04 12:49:32 +01001598 if (!pipe)
Jens Axboe912d35f2006-04-26 10:59:21 +02001599 return -EBADF;
Jens Axboe912d35f2006-04-26 10:59:21 +02001600
Jens Axboe35f3d142010-05-20 10:43:18 +02001601 if (splice_grow_spd(pipe, &spd))
1602 return -ENOMEM;
Jens Axboe912d35f2006-04-26 10:59:21 +02001603
Jens Axboe35f3d142010-05-20 10:43:18 +02001604 spd.nr_pages = get_iovec_page_array(iov, nr_segs, spd.pages,
Eric Dumazetbd1a68b2012-04-04 11:08:51 +02001605 spd.partial, false,
Eric Dumazet047fe362012-06-12 15:24:40 +02001606 spd.nr_pages_max);
Jens Axboe35f3d142010-05-20 10:43:18 +02001607 if (spd.nr_pages <= 0)
1608 ret = spd.nr_pages;
1609 else
1610 ret = splice_to_pipe(pipe, &spd);
1611
Eric Dumazet047fe362012-06-12 15:24:40 +02001612 splice_shrink_spd(&spd);
Jens Axboe35f3d142010-05-20 10:43:18 +02001613 return ret;
Jens Axboe912d35f2006-04-26 10:59:21 +02001614}
1615
Jens Axboe6a14b902007-06-14 13:08:55 +02001616/*
1617 * Note that vmsplice only really supports true splicing _from_ user memory
1618 * to a pipe, not the other way around. Splicing from user memory is a simple
1619 * operation that can be supported without any funky alignment restrictions
1620 * or nasty vm tricks. We simply map in the user memory and fill them into
1621 * a pipe. The reverse isn't quite as easy, though. There are two possible
1622 * solutions for that:
1623 *
1624 * - memcpy() the data internally, at which point we might as well just
1625 * do a regular read() on the buffer anyway.
1626 * - Lots of nasty vm tricks, that are neither fast nor flexible (it
1627 * has restriction limitations on both ends of the pipe).
1628 *
1629 * Currently we punt and implement it as a normal copy, see pipe_to_user().
1630 *
1631 */
Heiko Carstens836f92a2009-01-14 14:14:33 +01001632SYSCALL_DEFINE4(vmsplice, int, fd, const struct iovec __user *, iov,
1633 unsigned long, nr_segs, unsigned int, flags)
Jens Axboe912d35f2006-04-26 10:59:21 +02001634{
Al Viro2903ff02012-08-28 12:52:22 -04001635 struct fd f;
Jens Axboe912d35f2006-04-26 10:59:21 +02001636 long error;
Jens Axboe912d35f2006-04-26 10:59:21 +02001637
Jens Axboe6a14b902007-06-14 13:08:55 +02001638 if (unlikely(nr_segs > UIO_MAXIOV))
1639 return -EINVAL;
1640 else if (unlikely(!nr_segs))
1641 return 0;
1642
Jens Axboe912d35f2006-04-26 10:59:21 +02001643 error = -EBADF;
Al Viro2903ff02012-08-28 12:52:22 -04001644 f = fdget(fd);
1645 if (f.file) {
1646 if (f.file->f_mode & FMODE_WRITE)
1647 error = vmsplice_to_pipe(f.file, iov, nr_segs, flags);
1648 else if (f.file->f_mode & FMODE_READ)
1649 error = vmsplice_to_user(f.file, iov, nr_segs, flags);
Jens Axboe912d35f2006-04-26 10:59:21 +02001650
Al Viro2903ff02012-08-28 12:52:22 -04001651 fdput(f);
Jens Axboe912d35f2006-04-26 10:59:21 +02001652 }
1653
1654 return error;
1655}
1656
Al Viro76b021d2013-03-02 10:19:56 -05001657#ifdef CONFIG_COMPAT
1658COMPAT_SYSCALL_DEFINE4(vmsplice, int, fd, const struct compat_iovec __user *, iov32,
1659 unsigned int, nr_segs, unsigned int, flags)
1660{
1661 unsigned i;
1662 struct iovec __user *iov;
1663 if (nr_segs > UIO_MAXIOV)
1664 return -EINVAL;
1665 iov = compat_alloc_user_space(nr_segs * sizeof(struct iovec));
1666 for (i = 0; i < nr_segs; i++) {
1667 struct compat_iovec v;
1668 if (get_user(v.iov_base, &iov32[i].iov_base) ||
1669 get_user(v.iov_len, &iov32[i].iov_len) ||
1670 put_user(compat_ptr(v.iov_base), &iov[i].iov_base) ||
1671 put_user(v.iov_len, &iov[i].iov_len))
1672 return -EFAULT;
1673 }
1674 return sys_vmsplice(fd, iov, nr_segs, flags);
1675}
1676#endif
1677
Heiko Carstens836f92a2009-01-14 14:14:33 +01001678SYSCALL_DEFINE6(splice, int, fd_in, loff_t __user *, off_in,
1679 int, fd_out, loff_t __user *, off_out,
1680 size_t, len, unsigned int, flags)
Jens Axboe5274f052006-03-30 15:15:30 +02001681{
Al Viro2903ff02012-08-28 12:52:22 -04001682 struct fd in, out;
Jens Axboe5274f052006-03-30 15:15:30 +02001683 long error;
Jens Axboe5274f052006-03-30 15:15:30 +02001684
1685 if (unlikely(!len))
1686 return 0;
1687
1688 error = -EBADF;
Al Viro2903ff02012-08-28 12:52:22 -04001689 in = fdget(fd_in);
1690 if (in.file) {
1691 if (in.file->f_mode & FMODE_READ) {
1692 out = fdget(fd_out);
1693 if (out.file) {
1694 if (out.file->f_mode & FMODE_WRITE)
1695 error = do_splice(in.file, off_in,
1696 out.file, off_out,
Ingo Molnar529565d2006-04-10 15:18:58 +02001697 len, flags);
Al Viro2903ff02012-08-28 12:52:22 -04001698 fdput(out);
Jens Axboe5274f052006-03-30 15:15:30 +02001699 }
1700 }
Al Viro2903ff02012-08-28 12:52:22 -04001701 fdput(in);
Jens Axboe5274f052006-03-30 15:15:30 +02001702 }
Jens Axboe5274f052006-03-30 15:15:30 +02001703 return error;
1704}
Jens Axboe70524492006-04-11 15:51:17 +02001705
1706/*
Jens Axboeaadd06e2006-07-10 11:00:01 +02001707 * Make sure there's data to read. Wait for input if we can, otherwise
1708 * return an appropriate error.
1709 */
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001710static int ipipe_prep(struct pipe_inode_info *pipe, unsigned int flags)
Jens Axboeaadd06e2006-07-10 11:00:01 +02001711{
1712 int ret;
1713
1714 /*
1715 * Check ->nrbufs without the inode lock first. This function
1716 * is speculative anyways, so missing one is ok.
1717 */
1718 if (pipe->nrbufs)
1719 return 0;
1720
1721 ret = 0;
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001722 pipe_lock(pipe);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001723
1724 while (!pipe->nrbufs) {
1725 if (signal_pending(current)) {
1726 ret = -ERESTARTSYS;
1727 break;
1728 }
1729 if (!pipe->writers)
1730 break;
1731 if (!pipe->waiting_writers) {
1732 if (flags & SPLICE_F_NONBLOCK) {
1733 ret = -EAGAIN;
1734 break;
1735 }
1736 }
1737 pipe_wait(pipe);
1738 }
1739
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001740 pipe_unlock(pipe);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001741 return ret;
1742}
1743
1744/*
1745 * Make sure there's writeable room. Wait for room if we can, otherwise
1746 * return an appropriate error.
1747 */
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001748static int opipe_prep(struct pipe_inode_info *pipe, unsigned int flags)
Jens Axboeaadd06e2006-07-10 11:00:01 +02001749{
1750 int ret;
1751
1752 /*
1753 * Check ->nrbufs without the inode lock first. This function
1754 * is speculative anyways, so missing one is ok.
1755 */
Jens Axboe35f3d142010-05-20 10:43:18 +02001756 if (pipe->nrbufs < pipe->buffers)
Jens Axboeaadd06e2006-07-10 11:00:01 +02001757 return 0;
1758
1759 ret = 0;
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001760 pipe_lock(pipe);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001761
Jens Axboe35f3d142010-05-20 10:43:18 +02001762 while (pipe->nrbufs >= pipe->buffers) {
Jens Axboeaadd06e2006-07-10 11:00:01 +02001763 if (!pipe->readers) {
1764 send_sig(SIGPIPE, current, 0);
1765 ret = -EPIPE;
1766 break;
1767 }
1768 if (flags & SPLICE_F_NONBLOCK) {
1769 ret = -EAGAIN;
1770 break;
1771 }
1772 if (signal_pending(current)) {
1773 ret = -ERESTARTSYS;
1774 break;
1775 }
1776 pipe->waiting_writers++;
1777 pipe_wait(pipe);
1778 pipe->waiting_writers--;
1779 }
1780
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001781 pipe_unlock(pipe);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001782 return ret;
1783}
1784
1785/*
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001786 * Splice contents of ipipe to opipe.
1787 */
1788static int splice_pipe_to_pipe(struct pipe_inode_info *ipipe,
1789 struct pipe_inode_info *opipe,
1790 size_t len, unsigned int flags)
1791{
1792 struct pipe_buffer *ibuf, *obuf;
1793 int ret = 0, nbuf;
1794 bool input_wakeup = false;
1795
1796
1797retry:
1798 ret = ipipe_prep(ipipe, flags);
1799 if (ret)
1800 return ret;
1801
1802 ret = opipe_prep(opipe, flags);
1803 if (ret)
1804 return ret;
1805
1806 /*
1807 * Potential ABBA deadlock, work around it by ordering lock
1808 * grabbing by pipe info address. Otherwise two different processes
1809 * could deadlock (one doing tee from A -> B, the other from B -> A).
1810 */
1811 pipe_double_lock(ipipe, opipe);
1812
1813 do {
1814 if (!opipe->readers) {
1815 send_sig(SIGPIPE, current, 0);
1816 if (!ret)
1817 ret = -EPIPE;
1818 break;
1819 }
1820
1821 if (!ipipe->nrbufs && !ipipe->writers)
1822 break;
1823
1824 /*
1825 * Cannot make any progress, because either the input
1826 * pipe is empty or the output pipe is full.
1827 */
Jens Axboe35f3d142010-05-20 10:43:18 +02001828 if (!ipipe->nrbufs || opipe->nrbufs >= opipe->buffers) {
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001829 /* Already processed some buffers, break */
1830 if (ret)
1831 break;
1832
1833 if (flags & SPLICE_F_NONBLOCK) {
1834 ret = -EAGAIN;
1835 break;
1836 }
1837
1838 /*
1839 * We raced with another reader/writer and haven't
1840 * managed to process any buffers. A zero return
1841 * value means EOF, so retry instead.
1842 */
1843 pipe_unlock(ipipe);
1844 pipe_unlock(opipe);
1845 goto retry;
1846 }
1847
1848 ibuf = ipipe->bufs + ipipe->curbuf;
Jens Axboe35f3d142010-05-20 10:43:18 +02001849 nbuf = (opipe->curbuf + opipe->nrbufs) & (opipe->buffers - 1);
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001850 obuf = opipe->bufs + nbuf;
1851
1852 if (len >= ibuf->len) {
1853 /*
1854 * Simply move the whole buffer from ipipe to opipe
1855 */
1856 *obuf = *ibuf;
1857 ibuf->ops = NULL;
1858 opipe->nrbufs++;
Jens Axboe35f3d142010-05-20 10:43:18 +02001859 ipipe->curbuf = (ipipe->curbuf + 1) & (ipipe->buffers - 1);
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001860 ipipe->nrbufs--;
1861 input_wakeup = true;
1862 } else {
1863 /*
1864 * Get a reference to this pipe buffer,
1865 * so we can copy the contents over.
1866 */
1867 ibuf->ops->get(ipipe, ibuf);
1868 *obuf = *ibuf;
1869
1870 /*
1871 * Don't inherit the gift flag, we need to
1872 * prevent multiple steals of this page.
1873 */
1874 obuf->flags &= ~PIPE_BUF_FLAG_GIFT;
1875
1876 obuf->len = len;
1877 opipe->nrbufs++;
1878 ibuf->offset += obuf->len;
1879 ibuf->len -= obuf->len;
1880 }
1881 ret += obuf->len;
1882 len -= obuf->len;
1883 } while (len);
1884
1885 pipe_unlock(ipipe);
1886 pipe_unlock(opipe);
1887
1888 /*
1889 * If we put data in the output pipe, wakeup any potential readers.
1890 */
Namhyung Kim825cdcb2011-05-23 19:58:53 +02001891 if (ret > 0)
1892 wakeup_pipe_readers(opipe);
1893
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001894 if (input_wakeup)
1895 wakeup_pipe_writers(ipipe);
1896
1897 return ret;
1898}
1899
1900/*
Jens Axboe70524492006-04-11 15:51:17 +02001901 * Link contents of ipipe to opipe.
1902 */
1903static int link_pipe(struct pipe_inode_info *ipipe,
1904 struct pipe_inode_info *opipe,
1905 size_t len, unsigned int flags)
1906{
1907 struct pipe_buffer *ibuf, *obuf;
Jens Axboeaadd06e2006-07-10 11:00:01 +02001908 int ret = 0, i = 0, nbuf;
Jens Axboe70524492006-04-11 15:51:17 +02001909
1910 /*
1911 * Potential ABBA deadlock, work around it by ordering lock
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001912 * grabbing by pipe info address. Otherwise two different processes
Jens Axboe70524492006-04-11 15:51:17 +02001913 * could deadlock (one doing tee from A -> B, the other from B -> A).
1914 */
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001915 pipe_double_lock(ipipe, opipe);
Jens Axboe70524492006-04-11 15:51:17 +02001916
Jens Axboeaadd06e2006-07-10 11:00:01 +02001917 do {
Jens Axboe70524492006-04-11 15:51:17 +02001918 if (!opipe->readers) {
1919 send_sig(SIGPIPE, current, 0);
1920 if (!ret)
1921 ret = -EPIPE;
1922 break;
1923 }
Jens Axboe70524492006-04-11 15:51:17 +02001924
1925 /*
Jens Axboeaadd06e2006-07-10 11:00:01 +02001926 * If we have iterated all input buffers or ran out of
1927 * output room, break.
Jens Axboe70524492006-04-11 15:51:17 +02001928 */
Jens Axboe35f3d142010-05-20 10:43:18 +02001929 if (i >= ipipe->nrbufs || opipe->nrbufs >= opipe->buffers)
Jens Axboe70524492006-04-11 15:51:17 +02001930 break;
Jens Axboeaadd06e2006-07-10 11:00:01 +02001931
Jens Axboe35f3d142010-05-20 10:43:18 +02001932 ibuf = ipipe->bufs + ((ipipe->curbuf + i) & (ipipe->buffers-1));
1933 nbuf = (opipe->curbuf + opipe->nrbufs) & (opipe->buffers - 1);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001934
Jens Axboe2a27250e2006-04-19 15:56:40 +02001935 /*
Jens Axboeaadd06e2006-07-10 11:00:01 +02001936 * Get a reference to this pipe buffer,
1937 * so we can copy the contents over.
Jens Axboe2a27250e2006-04-19 15:56:40 +02001938 */
Jens Axboeaadd06e2006-07-10 11:00:01 +02001939 ibuf->ops->get(ipipe, ibuf);
Jens Axboe70524492006-04-11 15:51:17 +02001940
Jens Axboeaadd06e2006-07-10 11:00:01 +02001941 obuf = opipe->bufs + nbuf;
1942 *obuf = *ibuf;
Jens Axboe70524492006-04-11 15:51:17 +02001943
Jens Axboeaadd06e2006-07-10 11:00:01 +02001944 /*
1945 * Don't inherit the gift flag, we need to
1946 * prevent multiple steals of this page.
1947 */
1948 obuf->flags &= ~PIPE_BUF_FLAG_GIFT;
1949
1950 if (obuf->len > len)
1951 obuf->len = len;
1952
1953 opipe->nrbufs++;
1954 ret += obuf->len;
1955 len -= obuf->len;
1956 i++;
1957 } while (len);
Jens Axboe70524492006-04-11 15:51:17 +02001958
Jens Axboe02cf01a2008-02-20 10:34:51 +01001959 /*
1960 * return EAGAIN if we have the potential of some data in the
1961 * future, otherwise just return 0
1962 */
1963 if (!ret && ipipe->waiting_writers && (flags & SPLICE_F_NONBLOCK))
1964 ret = -EAGAIN;
1965
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001966 pipe_unlock(ipipe);
1967 pipe_unlock(opipe);
Jens Axboe70524492006-04-11 15:51:17 +02001968
Jens Axboeaadd06e2006-07-10 11:00:01 +02001969 /*
1970 * If we put data in the output pipe, wakeup any potential readers.
1971 */
Namhyung Kim825cdcb2011-05-23 19:58:53 +02001972 if (ret > 0)
1973 wakeup_pipe_readers(opipe);
Jens Axboe70524492006-04-11 15:51:17 +02001974
1975 return ret;
1976}
1977
1978/*
1979 * This is a tee(1) implementation that works on pipes. It doesn't copy
1980 * any data, it simply references the 'in' pages on the 'out' pipe.
1981 * The 'flags' used are the SPLICE_F_* variants, currently the only
1982 * applicable one is SPLICE_F_NONBLOCK.
1983 */
1984static long do_tee(struct file *in, struct file *out, size_t len,
1985 unsigned int flags)
1986{
Linus Torvalds71993e62010-11-28 13:56:09 -08001987 struct pipe_inode_info *ipipe = get_pipe_info(in);
1988 struct pipe_inode_info *opipe = get_pipe_info(out);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001989 int ret = -EINVAL;
Jens Axboe70524492006-04-11 15:51:17 +02001990
1991 /*
Jens Axboeaadd06e2006-07-10 11:00:01 +02001992 * Duplicate the contents of ipipe to opipe without actually
1993 * copying the data.
Jens Axboe70524492006-04-11 15:51:17 +02001994 */
Jens Axboeaadd06e2006-07-10 11:00:01 +02001995 if (ipipe && opipe && ipipe != opipe) {
1996 /*
1997 * Keep going, unless we encounter an error. The ipipe/opipe
1998 * ordering doesn't really matter.
1999 */
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02002000 ret = ipipe_prep(ipipe, flags);
Jens Axboeaadd06e2006-07-10 11:00:01 +02002001 if (!ret) {
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02002002 ret = opipe_prep(opipe, flags);
Jens Axboe02cf01a2008-02-20 10:34:51 +01002003 if (!ret)
Jens Axboeaadd06e2006-07-10 11:00:01 +02002004 ret = link_pipe(ipipe, opipe, len, flags);
Jens Axboeaadd06e2006-07-10 11:00:01 +02002005 }
2006 }
Jens Axboe70524492006-04-11 15:51:17 +02002007
Jens Axboeaadd06e2006-07-10 11:00:01 +02002008 return ret;
Jens Axboe70524492006-04-11 15:51:17 +02002009}
2010
Heiko Carstens836f92a2009-01-14 14:14:33 +01002011SYSCALL_DEFINE4(tee, int, fdin, int, fdout, size_t, len, unsigned int, flags)
Jens Axboe70524492006-04-11 15:51:17 +02002012{
Al Viro2903ff02012-08-28 12:52:22 -04002013 struct fd in;
2014 int error;
Jens Axboe70524492006-04-11 15:51:17 +02002015
2016 if (unlikely(!len))
2017 return 0;
2018
2019 error = -EBADF;
Al Viro2903ff02012-08-28 12:52:22 -04002020 in = fdget(fdin);
2021 if (in.file) {
2022 if (in.file->f_mode & FMODE_READ) {
2023 struct fd out = fdget(fdout);
2024 if (out.file) {
2025 if (out.file->f_mode & FMODE_WRITE)
2026 error = do_tee(in.file, out.file,
2027 len, flags);
2028 fdput(out);
Jens Axboe70524492006-04-11 15:51:17 +02002029 }
2030 }
Al Viro2903ff02012-08-28 12:52:22 -04002031 fdput(in);
Jens Axboe70524492006-04-11 15:51:17 +02002032 }
2033
2034 return error;
2035}