blob: 82bc0d64fc38d538b6482adf7d5d279318405069 [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);
Abhi Das90330e62015-12-18 14:11:36 -0600418retry_lookup:
Miklos Szeredi32502b82008-07-04 09:35:17 +0200419 page = find_or_create_page(mapping, index,
420 mapping_gfp_mask(mapping));
421
422 if (!page) {
423 error = -ENOMEM;
424 break;
425 }
Jens Axboe35f3d142010-05-20 10:43:18 +0200426 page_cache_release(spd.pages[page_nr]);
427 spd.pages[page_nr] = page;
Jens Axboe7480a902006-04-11 13:52:47 +0200428 }
429 /*
430 * page was already under io and is now done, great
431 */
432 if (PageUptodate(page)) {
433 unlock_page(page);
434 goto fill_it;
435 }
436
Jens Axboe7480a902006-04-11 13:52:47 +0200437 /*
438 * need to read in the page
439 */
440 error = mapping->a_ops->readpage(in, page);
Jens Axboe7480a902006-04-11 13:52:47 +0200441 if (unlikely(error)) {
Jens Axboeeb207962006-04-27 11:05:22 +0200442 /*
Abhi Das90330e62015-12-18 14:11:36 -0600443 * Re-lookup the page
Jens Axboeeb207962006-04-27 11:05:22 +0200444 */
Jens Axboe7480a902006-04-11 13:52:47 +0200445 if (error == AOP_TRUNCATED_PAGE)
Abhi Das90330e62015-12-18 14:11:36 -0600446 goto retry_lookup;
Jens Axboeeb207962006-04-27 11:05:22 +0200447
Jens Axboe7480a902006-04-11 13:52:47 +0200448 break;
449 }
Jens Axboe620a3242007-06-07 09:39:42 +0200450 }
451fill_it:
452 /*
453 * i_size must be checked after PageUptodate.
454 */
455 isize = i_size_read(mapping->host);
456 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
457 if (unlikely(!isize || index > end_index))
458 break;
459
460 /*
461 * if this is the last page, see if we need to shrink
462 * the length and stop
463 */
464 if (end_index == index) {
465 unsigned int plen;
Jens Axboe91ad66e2006-04-19 15:55:10 +0200466
467 /*
Jens Axboe620a3242007-06-07 09:39:42 +0200468 * max good bytes in this page
Jens Axboe91ad66e2006-04-19 15:55:10 +0200469 */
Jens Axboe620a3242007-06-07 09:39:42 +0200470 plen = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
471 if (plen <= loff)
Jens Axboe91ad66e2006-04-19 15:55:10 +0200472 break;
Jens Axboe91ad66e2006-04-19 15:55:10 +0200473
474 /*
Jens Axboe620a3242007-06-07 09:39:42 +0200475 * force quit after adding this page
Jens Axboe91ad66e2006-04-19 15:55:10 +0200476 */
Jens Axboe620a3242007-06-07 09:39:42 +0200477 this_len = min(this_len, plen - loff);
478 len = this_len;
Jens Axboe7480a902006-04-11 13:52:47 +0200479 }
Jens Axboe620a3242007-06-07 09:39:42 +0200480
Jens Axboe35f3d142010-05-20 10:43:18 +0200481 spd.partial[page_nr].offset = loff;
482 spd.partial[page_nr].len = this_len;
Jens Axboe82aa5d62006-04-20 13:05:48 +0200483 len -= this_len;
Jens Axboe91ad66e2006-04-19 15:55:10 +0200484 loff = 0;
Jens Axboeeb207962006-04-27 11:05:22 +0200485 spd.nr_pages++;
486 index++;
Jens Axboe5274f052006-03-30 15:15:30 +0200487 }
488
Jens Axboeeb207962006-04-27 11:05:22 +0200489 /*
Hugh Dickins475ecad2007-06-07 09:36:00 +0200490 * Release any pages at the end, if we quit early. 'page_nr' is how far
Jens Axboeeb207962006-04-27 11:05:22 +0200491 * we got, 'nr_pages' is how many pages are in the map.
492 */
493 while (page_nr < nr_pages)
Jens Axboe35f3d142010-05-20 10:43:18 +0200494 page_cache_release(spd.pages[page_nr++]);
Fengguang Wuf4e6b492007-10-16 01:24:33 -0700495 in->f_ra.prev_pos = (loff_t)index << PAGE_CACHE_SHIFT;
Jens Axboeeb207962006-04-27 11:05:22 +0200496
Jens Axboe912d35f2006-04-26 10:59:21 +0200497 if (spd.nr_pages)
Jens Axboe35f3d142010-05-20 10:43:18 +0200498 error = splice_to_pipe(pipe, &spd);
Jens Axboe5274f052006-03-30 15:15:30 +0200499
Eric Dumazet047fe362012-06-12 15:24:40 +0200500 splice_shrink_spd(&spd);
Jens Axboe7480a902006-04-11 13:52:47 +0200501 return error;
Jens Axboe5274f052006-03-30 15:15:30 +0200502}
503
Jens Axboe83f91352006-04-02 23:05:09 +0200504/**
505 * generic_file_splice_read - splice data from file to a pipe
506 * @in: file to splice from
Jens Axboe932cc6d2007-06-21 13:10:21 +0200507 * @ppos: position in @in
Jens Axboe83f91352006-04-02 23:05:09 +0200508 * @pipe: pipe to splice to
509 * @len: number of bytes to splice
510 * @flags: splice modifier flags
511 *
Jens Axboe932cc6d2007-06-21 13:10:21 +0200512 * Description:
513 * Will read pages from given file and fill them into a pipe. Can be
514 * used as long as the address_space operations for the source implements
515 * a readpage() hook.
516 *
Jens Axboe83f91352006-04-02 23:05:09 +0200517 */
Jens Axboecbb7e572006-04-11 14:57:50 +0200518ssize_t generic_file_splice_read(struct file *in, loff_t *ppos,
519 struct pipe_inode_info *pipe, size_t len,
520 unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +0200521{
Jens Axboed366d3982007-06-01 14:54:11 +0200522 loff_t isize, left;
Jens Axboe8191ecd2008-04-10 08:24:25 +0200523 int ret;
Jens Axboed366d3982007-06-01 14:54:11 +0200524
Boaz Harroshbe64f882015-04-15 16:15:17 -0700525 if (IS_DAX(in->f_mapping->host))
526 return default_file_splice_read(in, ppos, pipe, len, flags);
527
Jens Axboed366d3982007-06-01 14:54:11 +0200528 isize = i_size_read(in->f_mapping->host);
529 if (unlikely(*ppos >= isize))
530 return 0;
531
532 left = isize - *ppos;
533 if (unlikely(left < len))
534 len = left;
Jens Axboe5274f052006-03-30 15:15:30 +0200535
Jens Axboe8191ecd2008-04-10 08:24:25 +0200536 ret = __generic_file_splice_read(in, ppos, pipe, len, flags);
Miklos Szeredi723590e2009-08-15 08:43:22 +0200537 if (ret > 0) {
Jens Axboecbb7e572006-04-11 14:57:50 +0200538 *ppos += ret;
Miklos Szeredi723590e2009-08-15 08:43:22 +0200539 file_accessed(in);
540 }
Jens Axboe5274f052006-03-30 15:15:30 +0200541
542 return ret;
543}
Jens Axboe059a8f32006-04-02 23:06:05 +0200544EXPORT_SYMBOL(generic_file_splice_read);
545
Miklos Szeredi68181732009-05-07 15:37:36 +0200546static const struct pipe_buf_operations default_pipe_buf_ops = {
547 .can_merge = 0,
Miklos Szeredi68181732009-05-07 15:37:36 +0200548 .confirm = generic_pipe_buf_confirm,
549 .release = generic_pipe_buf_release,
550 .steal = generic_pipe_buf_steal,
551 .get = generic_pipe_buf_get,
552};
553
Miklos Szeredi28a625c2014-01-22 19:36:57 +0100554static int generic_pipe_buf_nosteal(struct pipe_inode_info *pipe,
555 struct pipe_buffer *buf)
556{
557 return 1;
558}
559
560/* Pipe buffer operations for a socket and similar. */
561const struct pipe_buf_operations nosteal_pipe_buf_ops = {
562 .can_merge = 0,
Miklos Szeredi28a625c2014-01-22 19:36:57 +0100563 .confirm = generic_pipe_buf_confirm,
564 .release = generic_pipe_buf_release,
565 .steal = generic_pipe_buf_nosteal,
566 .get = generic_pipe_buf_get,
567};
568EXPORT_SYMBOL(nosteal_pipe_buf_ops);
569
Miklos Szeredi68181732009-05-07 15:37:36 +0200570static ssize_t kernel_readv(struct file *file, const struct iovec *vec,
571 unsigned long vlen, loff_t offset)
572{
573 mm_segment_t old_fs;
574 loff_t pos = offset;
575 ssize_t res;
576
577 old_fs = get_fs();
578 set_fs(get_ds());
579 /* The cast to a user pointer is valid due to the set_fs() */
580 res = vfs_readv(file, (const struct iovec __user *)vec, vlen, &pos);
581 set_fs(old_fs);
582
583 return res;
584}
585
Al Viro7bb307e2013-02-23 14:51:48 -0500586ssize_t kernel_write(struct file *file, const char *buf, size_t count,
Miklos Szeredib2858d72009-05-19 11:37:46 +0200587 loff_t pos)
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +0200588{
589 mm_segment_t old_fs;
590 ssize_t res;
591
592 old_fs = get_fs();
593 set_fs(get_ds());
594 /* The cast to a user pointer is valid due to the set_fs() */
Al Viro7bb307e2013-02-23 14:51:48 -0500595 res = vfs_write(file, (__force const char __user *)buf, count, &pos);
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +0200596 set_fs(old_fs);
597
598 return res;
599}
Al Viro7bb307e2013-02-23 14:51:48 -0500600EXPORT_SYMBOL(kernel_write);
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +0200601
Miklos Szeredi68181732009-05-07 15:37:36 +0200602ssize_t default_file_splice_read(struct file *in, loff_t *ppos,
603 struct pipe_inode_info *pipe, size_t len,
604 unsigned int flags)
605{
606 unsigned int nr_pages;
607 unsigned int nr_freed;
608 size_t offset;
Jens Axboe35f3d142010-05-20 10:43:18 +0200609 struct page *pages[PIPE_DEF_BUFFERS];
610 struct partial_page partial[PIPE_DEF_BUFFERS];
611 struct iovec *vec, __vec[PIPE_DEF_BUFFERS];
Miklos Szeredi68181732009-05-07 15:37:36 +0200612 ssize_t res;
613 size_t this_len;
614 int error;
615 int i;
616 struct splice_pipe_desc spd = {
617 .pages = pages,
618 .partial = partial,
Eric Dumazet047fe362012-06-12 15:24:40 +0200619 .nr_pages_max = PIPE_DEF_BUFFERS,
Miklos Szeredi68181732009-05-07 15:37:36 +0200620 .flags = flags,
621 .ops = &default_pipe_buf_ops,
622 .spd_release = spd_release_page,
623 };
624
Jens Axboe35f3d142010-05-20 10:43:18 +0200625 if (splice_grow_spd(pipe, &spd))
626 return -ENOMEM;
627
628 res = -ENOMEM;
629 vec = __vec;
Eric Dumazet047fe362012-06-12 15:24:40 +0200630 if (spd.nr_pages_max > PIPE_DEF_BUFFERS) {
631 vec = kmalloc(spd.nr_pages_max * sizeof(struct iovec), GFP_KERNEL);
Jens Axboe35f3d142010-05-20 10:43:18 +0200632 if (!vec)
633 goto shrink_ret;
634 }
635
Miklos Szeredi68181732009-05-07 15:37:36 +0200636 offset = *ppos & ~PAGE_CACHE_MASK;
637 nr_pages = (len + offset + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
638
Eric Dumazet047fe362012-06-12 15:24:40 +0200639 for (i = 0; i < nr_pages && i < spd.nr_pages_max && len; i++) {
Miklos Szeredi68181732009-05-07 15:37:36 +0200640 struct page *page;
641
Jens Axboe4f231222009-05-13 08:35:35 +0200642 page = alloc_page(GFP_USER);
Miklos Szeredi68181732009-05-07 15:37:36 +0200643 error = -ENOMEM;
644 if (!page)
645 goto err;
646
647 this_len = min_t(size_t, len, PAGE_CACHE_SIZE - offset);
Jens Axboe4f231222009-05-13 08:35:35 +0200648 vec[i].iov_base = (void __user *) page_address(page);
Miklos Szeredi68181732009-05-07 15:37:36 +0200649 vec[i].iov_len = this_len;
Jens Axboe35f3d142010-05-20 10:43:18 +0200650 spd.pages[i] = page;
Miklos Szeredi68181732009-05-07 15:37:36 +0200651 spd.nr_pages++;
652 len -= this_len;
653 offset = 0;
654 }
655
656 res = kernel_readv(in, vec, spd.nr_pages, *ppos);
Andrew Morton77f6bf52009-05-14 09:49:44 +0200657 if (res < 0) {
658 error = res;
Miklos Szeredi68181732009-05-07 15:37:36 +0200659 goto err;
Andrew Morton77f6bf52009-05-14 09:49:44 +0200660 }
Miklos Szeredi68181732009-05-07 15:37:36 +0200661
662 error = 0;
663 if (!res)
664 goto err;
665
666 nr_freed = 0;
667 for (i = 0; i < spd.nr_pages; i++) {
Miklos Szeredi68181732009-05-07 15:37:36 +0200668 this_len = min_t(size_t, vec[i].iov_len, res);
Jens Axboe35f3d142010-05-20 10:43:18 +0200669 spd.partial[i].offset = 0;
670 spd.partial[i].len = this_len;
Miklos Szeredi68181732009-05-07 15:37:36 +0200671 if (!this_len) {
Jens Axboe35f3d142010-05-20 10:43:18 +0200672 __free_page(spd.pages[i]);
673 spd.pages[i] = NULL;
Miklos Szeredi68181732009-05-07 15:37:36 +0200674 nr_freed++;
675 }
676 res -= this_len;
677 }
678 spd.nr_pages -= nr_freed;
679
680 res = splice_to_pipe(pipe, &spd);
681 if (res > 0)
682 *ppos += res;
683
Jens Axboe35f3d142010-05-20 10:43:18 +0200684shrink_ret:
685 if (vec != __vec)
686 kfree(vec);
Eric Dumazet047fe362012-06-12 15:24:40 +0200687 splice_shrink_spd(&spd);
Miklos Szeredi68181732009-05-07 15:37:36 +0200688 return res;
689
690err:
Jens Axboe4f231222009-05-13 08:35:35 +0200691 for (i = 0; i < spd.nr_pages; i++)
Jens Axboe35f3d142010-05-20 10:43:18 +0200692 __free_page(spd.pages[i]);
Jens Axboe4f231222009-05-13 08:35:35 +0200693
Jens Axboe35f3d142010-05-20 10:43:18 +0200694 res = error;
695 goto shrink_ret;
Miklos Szeredi68181732009-05-07 15:37:36 +0200696}
697EXPORT_SYMBOL(default_file_splice_read);
698
Jens Axboe5274f052006-03-30 15:15:30 +0200699/*
Jens Axboe4f6f0bd2006-04-02 23:04:46 +0200700 * Send 'sd->len' bytes to socket from 'sd->file' at position 'sd->pos'
Jens Axboe016b6612006-04-25 15:42:00 +0200701 * using sendpage(). Return the number of bytes sent.
Jens Axboe5274f052006-03-30 15:15:30 +0200702 */
Jens Axboe76ad4d12006-05-03 10:41:33 +0200703static int pipe_to_sendpage(struct pipe_inode_info *pipe,
Jens Axboe5274f052006-03-30 15:15:30 +0200704 struct pipe_buffer *buf, struct splice_desc *sd)
705{
Jens Axboe6a14b902007-06-14 13:08:55 +0200706 struct file *file = sd->u.file;
Jens Axboe5274f052006-03-30 15:15:30 +0200707 loff_t pos = sd->pos;
Michał Mirosława8adbe32010-12-17 08:56:44 +0100708 int more;
Jens Axboe5274f052006-03-30 15:15:30 +0200709
Al Viro72c2d532013-09-22 16:27:52 -0400710 if (!likely(file->f_op->sendpage))
Michał Mirosława8adbe32010-12-17 08:56:44 +0100711 return -EINVAL;
Jens Axboe5274f052006-03-30 15:15:30 +0200712
Eric Dumazet35f9c092012-04-05 03:05:35 +0000713 more = (sd->flags & SPLICE_F_MORE) ? MSG_MORE : 0;
Eric Dumazetae62ca72013-01-06 18:21:49 +0000714
715 if (sd->len < sd->total_len && pipe->nrbufs > 1)
Eric Dumazet35f9c092012-04-05 03:05:35 +0000716 more |= MSG_SENDPAGE_NOTLAST;
Eric Dumazetae62ca72013-01-06 18:21:49 +0000717
Michał Mirosława8adbe32010-12-17 08:56:44 +0100718 return file->f_op->sendpage(file, buf->page, buf->offset,
719 sd->len, &pos, more);
Jens Axboe5274f052006-03-30 15:15:30 +0200720}
721
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200722static void wakeup_pipe_writers(struct pipe_inode_info *pipe)
723{
724 smp_mb();
725 if (waitqueue_active(&pipe->wait))
726 wake_up_interruptible(&pipe->wait);
727 kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
728}
729
730/**
731 * splice_from_pipe_feed - feed available data from a pipe to a file
732 * @pipe: pipe to splice from
733 * @sd: information to @actor
734 * @actor: handler that splices the data
735 *
736 * Description:
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200737 * This function loops over the pipe and calls @actor to do the
738 * actual moving of a single struct pipe_buffer to the desired
739 * destination. It returns when there's no more buffers left in
740 * the pipe or if the requested number of bytes (@sd->total_len)
741 * have been copied. It returns a positive number (one) if the
742 * pipe needs to be filled with more data, zero if the required
743 * number of bytes have been copied and -errno on error.
744 *
745 * This, together with splice_from_pipe_{begin,end,next}, may be
746 * used to implement the functionality of __splice_from_pipe() when
747 * locking is required around copying the pipe buffers to the
748 * destination.
749 */
Al Viro96f9bc82014-04-05 04:35:49 -0400750static int splice_from_pipe_feed(struct pipe_inode_info *pipe, struct splice_desc *sd,
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200751 splice_actor *actor)
752{
753 int ret;
754
755 while (pipe->nrbufs) {
756 struct pipe_buffer *buf = pipe->bufs + pipe->curbuf;
757 const struct pipe_buf_operations *ops = buf->ops;
758
759 sd->len = buf->len;
760 if (sd->len > sd->total_len)
761 sd->len = sd->total_len;
762
Michał Mirosława8adbe32010-12-17 08:56:44 +0100763 ret = buf->ops->confirm(pipe, buf);
764 if (unlikely(ret)) {
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200765 if (ret == -ENODATA)
766 ret = 0;
767 return ret;
768 }
Michał Mirosława8adbe32010-12-17 08:56:44 +0100769
770 ret = actor(pipe, buf, sd);
771 if (ret <= 0)
772 return ret;
773
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200774 buf->offset += ret;
775 buf->len -= ret;
776
777 sd->num_spliced += ret;
778 sd->len -= ret;
779 sd->pos += ret;
780 sd->total_len -= ret;
781
782 if (!buf->len) {
783 buf->ops = NULL;
784 ops->release(pipe, buf);
Jens Axboe35f3d142010-05-20 10:43:18 +0200785 pipe->curbuf = (pipe->curbuf + 1) & (pipe->buffers - 1);
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200786 pipe->nrbufs--;
Al Viro6447a3c2013-03-21 11:01:38 -0400787 if (pipe->files)
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200788 sd->need_wakeup = true;
789 }
790
791 if (!sd->total_len)
792 return 0;
793 }
794
795 return 1;
796}
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200797
798/**
799 * splice_from_pipe_next - wait for some data to splice from
800 * @pipe: pipe to splice from
801 * @sd: information about the splice operation
802 *
803 * Description:
804 * This function will wait for some data and return a positive
805 * value (one) if pipe buffers are available. It will return zero
806 * or -errno if no more data needs to be spliced.
807 */
Al Viro96f9bc82014-04-05 04:35:49 -0400808static int splice_from_pipe_next(struct pipe_inode_info *pipe, struct splice_desc *sd)
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200809{
Jan Karac725bfc2015-11-23 13:09:50 +0100810 /*
811 * Check for signal early to make process killable when there are
812 * always buffers available
813 */
814 if (signal_pending(current))
815 return -ERESTARTSYS;
816
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200817 while (!pipe->nrbufs) {
818 if (!pipe->writers)
819 return 0;
820
821 if (!pipe->waiting_writers && sd->num_spliced)
822 return 0;
823
824 if (sd->flags & SPLICE_F_NONBLOCK)
825 return -EAGAIN;
826
827 if (signal_pending(current))
828 return -ERESTARTSYS;
829
830 if (sd->need_wakeup) {
831 wakeup_pipe_writers(pipe);
832 sd->need_wakeup = false;
833 }
834
835 pipe_wait(pipe);
836 }
837
838 return 1;
839}
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200840
841/**
842 * splice_from_pipe_begin - start splicing from pipe
Randy Dunlapb80901b2009-04-16 19:09:55 -0700843 * @sd: information about the splice operation
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200844 *
845 * Description:
846 * This function should be called before a loop containing
847 * splice_from_pipe_next() and splice_from_pipe_feed() to
848 * initialize the necessary fields of @sd.
849 */
Al Viro96f9bc82014-04-05 04:35:49 -0400850static void splice_from_pipe_begin(struct splice_desc *sd)
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200851{
852 sd->num_spliced = 0;
853 sd->need_wakeup = false;
854}
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200855
856/**
857 * splice_from_pipe_end - finish splicing from pipe
858 * @pipe: pipe to splice from
859 * @sd: information about the splice operation
860 *
861 * Description:
862 * This function will wake up pipe writers if necessary. It should
863 * be called after a loop containing splice_from_pipe_next() and
864 * splice_from_pipe_feed().
865 */
Al Viro96f9bc82014-04-05 04:35:49 -0400866static void splice_from_pipe_end(struct pipe_inode_info *pipe, struct splice_desc *sd)
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200867{
868 if (sd->need_wakeup)
869 wakeup_pipe_writers(pipe);
870}
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200871
Jens Axboe932cc6d2007-06-21 13:10:21 +0200872/**
873 * __splice_from_pipe - splice data from a pipe to given actor
874 * @pipe: pipe to splice from
875 * @sd: information to @actor
876 * @actor: handler that splices the data
877 *
878 * Description:
879 * This function does little more than loop over the pipe and call
880 * @actor to do the actual moving of a single struct pipe_buffer to
881 * the desired destination. See pipe_to_file, pipe_to_sendpage, or
882 * pipe_to_user.
883 *
Jens Axboe83f91352006-04-02 23:05:09 +0200884 */
Jens Axboec66ab6f2007-06-12 21:17:17 +0200885ssize_t __splice_from_pipe(struct pipe_inode_info *pipe, struct splice_desc *sd,
886 splice_actor *actor)
Jens Axboe5274f052006-03-30 15:15:30 +0200887{
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200888 int ret;
Jens Axboe5274f052006-03-30 15:15:30 +0200889
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200890 splice_from_pipe_begin(sd);
891 do {
Jan Karac2489e02015-11-23 13:09:51 +0100892 cond_resched();
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200893 ret = splice_from_pipe_next(pipe, sd);
894 if (ret > 0)
895 ret = splice_from_pipe_feed(pipe, sd, actor);
896 } while (ret > 0);
897 splice_from_pipe_end(pipe, sd);
Jens Axboe5274f052006-03-30 15:15:30 +0200898
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200899 return sd->num_spliced ? sd->num_spliced : ret;
Jens Axboe5274f052006-03-30 15:15:30 +0200900}
Mark Fasheh40bee44e2007-03-21 13:11:02 +0100901EXPORT_SYMBOL(__splice_from_pipe);
Jens Axboe5274f052006-03-30 15:15:30 +0200902
Jens Axboe932cc6d2007-06-21 13:10:21 +0200903/**
904 * splice_from_pipe - splice data from a pipe to a file
905 * @pipe: pipe to splice from
906 * @out: file to splice to
907 * @ppos: position in @out
908 * @len: how many bytes to splice
909 * @flags: splice modifier flags
910 * @actor: handler that splices the data
911 *
912 * Description:
Miklos Szeredi29339702009-04-14 19:48:37 +0200913 * See __splice_from_pipe. This function locks the pipe inode,
Jens Axboe932cc6d2007-06-21 13:10:21 +0200914 * otherwise it's identical to __splice_from_pipe().
915 *
916 */
Mark Fasheh6da61802006-10-17 18:43:07 +0200917ssize_t splice_from_pipe(struct pipe_inode_info *pipe, struct file *out,
918 loff_t *ppos, size_t len, unsigned int flags,
919 splice_actor *actor)
920{
921 ssize_t ret;
Jens Axboec66ab6f2007-06-12 21:17:17 +0200922 struct splice_desc sd = {
923 .total_len = len,
924 .flags = flags,
925 .pos = *ppos,
Jens Axboe6a14b902007-06-14 13:08:55 +0200926 .u.file = out,
Jens Axboec66ab6f2007-06-12 21:17:17 +0200927 };
Mark Fasheh6da61802006-10-17 18:43:07 +0200928
Miklos Szeredi61e0d472009-04-14 19:48:41 +0200929 pipe_lock(pipe);
Jens Axboec66ab6f2007-06-12 21:17:17 +0200930 ret = __splice_from_pipe(pipe, &sd, actor);
Miklos Szeredi61e0d472009-04-14 19:48:41 +0200931 pipe_unlock(pipe);
Mark Fasheh6da61802006-10-17 18:43:07 +0200932
933 return ret;
934}
935
936/**
Al Viro8d020762014-04-05 04:27:08 -0400937 * iter_file_splice_write - splice data from a pipe to a file
938 * @pipe: pipe info
939 * @out: file to write to
940 * @ppos: position in @out
941 * @len: number of bytes to splice
942 * @flags: splice modifier flags
943 *
944 * Description:
945 * Will either move or copy pages (determined by @flags options) from
946 * the given pipe inode to the given file.
947 * This one is ->write_iter-based.
948 *
949 */
950ssize_t
951iter_file_splice_write(struct pipe_inode_info *pipe, struct file *out,
952 loff_t *ppos, size_t len, unsigned int flags)
953{
954 struct splice_desc sd = {
955 .total_len = len,
956 .flags = flags,
957 .pos = *ppos,
958 .u.file = out,
959 };
960 int nbufs = pipe->buffers;
961 struct bio_vec *array = kcalloc(nbufs, sizeof(struct bio_vec),
962 GFP_KERNEL);
963 ssize_t ret;
964
965 if (unlikely(!array))
966 return -ENOMEM;
967
968 pipe_lock(pipe);
969
970 splice_from_pipe_begin(&sd);
971 while (sd.total_len) {
972 struct iov_iter from;
Al Viro8d020762014-04-05 04:27:08 -0400973 size_t left;
974 int n, idx;
975
976 ret = splice_from_pipe_next(pipe, &sd);
977 if (ret <= 0)
978 break;
979
980 if (unlikely(nbufs < pipe->buffers)) {
981 kfree(array);
982 nbufs = pipe->buffers;
983 array = kcalloc(nbufs, sizeof(struct bio_vec),
984 GFP_KERNEL);
985 if (!array) {
986 ret = -ENOMEM;
987 break;
988 }
989 }
990
991 /* build the vector */
992 left = sd.total_len;
993 for (n = 0, idx = pipe->curbuf; left && n < pipe->nrbufs; n++, idx++) {
994 struct pipe_buffer *buf = pipe->bufs + idx;
995 size_t this_len = buf->len;
996
997 if (this_len > left)
998 this_len = left;
999
1000 if (idx == pipe->buffers - 1)
1001 idx = -1;
1002
1003 ret = buf->ops->confirm(pipe, buf);
1004 if (unlikely(ret)) {
1005 if (ret == -ENODATA)
1006 ret = 0;
1007 goto done;
1008 }
1009
1010 array[n].bv_page = buf->page;
1011 array[n].bv_len = this_len;
1012 array[n].bv_offset = buf->offset;
1013 left -= this_len;
1014 }
1015
Al Viro05afcb72015-01-23 01:08:07 -05001016 iov_iter_bvec(&from, ITER_BVEC | WRITE, array, n,
1017 sd.total_len - left);
Christoph Hellwigdbe4e192015-01-25 21:11:59 +01001018 ret = vfs_iter_write(out, &from, &sd.pos);
Al Viro8d020762014-04-05 04:27:08 -04001019 if (ret <= 0)
1020 break;
1021
1022 sd.num_spliced += ret;
1023 sd.total_len -= ret;
Christoph Hellwigdbe4e192015-01-25 21:11:59 +01001024 *ppos = sd.pos;
Al Viro8d020762014-04-05 04:27:08 -04001025
1026 /* dismiss the fully eaten buffers, adjust the partial one */
1027 while (ret) {
1028 struct pipe_buffer *buf = pipe->bufs + pipe->curbuf;
1029 if (ret >= buf->len) {
1030 const struct pipe_buf_operations *ops = buf->ops;
1031 ret -= buf->len;
1032 buf->len = 0;
1033 buf->ops = NULL;
1034 ops->release(pipe, buf);
1035 pipe->curbuf = (pipe->curbuf + 1) & (pipe->buffers - 1);
1036 pipe->nrbufs--;
1037 if (pipe->files)
1038 sd.need_wakeup = true;
1039 } else {
1040 buf->offset += ret;
1041 buf->len -= ret;
1042 ret = 0;
1043 }
1044 }
1045 }
1046done:
1047 kfree(array);
1048 splice_from_pipe_end(pipe, &sd);
1049
1050 pipe_unlock(pipe);
1051
1052 if (sd.num_spliced)
1053 ret = sd.num_spliced;
1054
1055 return ret;
1056}
1057
1058EXPORT_SYMBOL(iter_file_splice_write);
1059
Miklos Szeredib2858d72009-05-19 11:37:46 +02001060static int write_pipe_buf(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
1061 struct splice_desc *sd)
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +02001062{
Miklos Szeredib2858d72009-05-19 11:37:46 +02001063 int ret;
1064 void *data;
Al Viro06ae43f2013-03-20 13:19:30 -04001065 loff_t tmp = sd->pos;
Miklos Szeredib2858d72009-05-19 11:37:46 +02001066
Al Virofbb32752014-02-02 21:09:54 -05001067 data = kmap(buf->page);
Al Viro06ae43f2013-03-20 13:19:30 -04001068 ret = __kernel_write(sd->u.file, data + buf->offset, sd->len, &tmp);
Al Virofbb32752014-02-02 21:09:54 -05001069 kunmap(buf->page);
Miklos Szeredib2858d72009-05-19 11:37:46 +02001070
1071 return ret;
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +02001072}
1073
1074static ssize_t default_file_splice_write(struct pipe_inode_info *pipe,
1075 struct file *out, loff_t *ppos,
1076 size_t len, unsigned int flags)
1077{
Miklos Szeredib2858d72009-05-19 11:37:46 +02001078 ssize_t ret;
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +02001079
Miklos Szeredib2858d72009-05-19 11:37:46 +02001080 ret = splice_from_pipe(pipe, out, ppos, len, flags, write_pipe_buf);
1081 if (ret > 0)
1082 *ppos += ret;
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +02001083
Miklos Szeredib2858d72009-05-19 11:37:46 +02001084 return ret;
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +02001085}
1086
Jens Axboe83f91352006-04-02 23:05:09 +02001087/**
1088 * generic_splice_sendpage - splice data from a pipe to a socket
Jens Axboe932cc6d2007-06-21 13:10:21 +02001089 * @pipe: pipe to splice from
Jens Axboe83f91352006-04-02 23:05:09 +02001090 * @out: socket to write to
Jens Axboe932cc6d2007-06-21 13:10:21 +02001091 * @ppos: position in @out
Jens Axboe83f91352006-04-02 23:05:09 +02001092 * @len: number of bytes to splice
1093 * @flags: splice modifier flags
1094 *
Jens Axboe932cc6d2007-06-21 13:10:21 +02001095 * Description:
1096 * Will send @len bytes from the pipe to a network socket. No data copying
1097 * is involved.
Jens Axboe83f91352006-04-02 23:05:09 +02001098 *
1099 */
Ingo Molnar3a326a22006-04-10 15:18:35 +02001100ssize_t generic_splice_sendpage(struct pipe_inode_info *pipe, struct file *out,
Jens Axboecbb7e572006-04-11 14:57:50 +02001101 loff_t *ppos, size_t len, unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +02001102{
Jens Axboe00522fb2006-04-26 14:39:29 +02001103 return splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_sendpage);
Jens Axboe5274f052006-03-30 15:15:30 +02001104}
1105
Jens Axboe059a8f32006-04-02 23:06:05 +02001106EXPORT_SYMBOL(generic_splice_sendpage);
Jeff Garzika0f06782006-03-30 23:06:13 -05001107
Jens Axboe83f91352006-04-02 23:05:09 +02001108/*
1109 * Attempt to initiate a splice from pipe to file.
1110 */
Ingo Molnar3a326a22006-04-10 15:18:35 +02001111static long do_splice_from(struct pipe_inode_info *pipe, struct file *out,
Jens Axboecbb7e572006-04-11 14:57:50 +02001112 loff_t *ppos, size_t len, unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +02001113{
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +02001114 ssize_t (*splice_write)(struct pipe_inode_info *, struct file *,
1115 loff_t *, size_t, unsigned int);
Jens Axboe5274f052006-03-30 15:15:30 +02001116
Al Viro72c2d532013-09-22 16:27:52 -04001117 if (out->f_op->splice_write)
Changli Gaocc56f7d2009-11-04 09:09:52 +01001118 splice_write = out->f_op->splice_write;
1119 else
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +02001120 splice_write = default_file_splice_write;
1121
Al Viro500368f2013-05-23 20:07:11 -04001122 return splice_write(pipe, out, ppos, len, flags);
Jens Axboe5274f052006-03-30 15:15:30 +02001123}
1124
Jens Axboe83f91352006-04-02 23:05:09 +02001125/*
1126 * Attempt to initiate a splice from a file to a pipe.
1127 */
Jens Axboecbb7e572006-04-11 14:57:50 +02001128static long do_splice_to(struct file *in, loff_t *ppos,
1129 struct pipe_inode_info *pipe, size_t len,
1130 unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +02001131{
Miklos Szeredi68181732009-05-07 15:37:36 +02001132 ssize_t (*splice_read)(struct file *, loff_t *,
1133 struct pipe_inode_info *, size_t, unsigned int);
Jens Axboe5274f052006-03-30 15:15:30 +02001134 int ret;
1135
Jens Axboe49570e92006-04-11 13:56:09 +02001136 if (unlikely(!(in->f_mode & FMODE_READ)))
Jens Axboe5274f052006-03-30 15:15:30 +02001137 return -EBADF;
1138
Jens Axboecbb7e572006-04-11 14:57:50 +02001139 ret = rw_verify_area(READ, in, ppos, len);
Jens Axboe5274f052006-03-30 15:15:30 +02001140 if (unlikely(ret < 0))
1141 return ret;
1142
Al Viro72c2d532013-09-22 16:27:52 -04001143 if (in->f_op->splice_read)
Changli Gaocc56f7d2009-11-04 09:09:52 +01001144 splice_read = in->f_op->splice_read;
1145 else
Miklos Szeredi68181732009-05-07 15:37:36 +02001146 splice_read = default_file_splice_read;
1147
1148 return splice_read(in, ppos, pipe, len, flags);
Jens Axboe5274f052006-03-30 15:15:30 +02001149}
1150
Jens Axboe932cc6d2007-06-21 13:10:21 +02001151/**
1152 * splice_direct_to_actor - splices data directly between two non-pipes
1153 * @in: file to splice from
1154 * @sd: actor information on where to splice to
1155 * @actor: handles the data splicing
1156 *
1157 * Description:
1158 * This is a special case helper to splice directly between two
1159 * points, without requiring an explicit pipe. Internally an allocated
Randy Dunlap79685b82007-07-27 08:08:51 +02001160 * pipe is cached in the process, and reused during the lifetime of
Jens Axboe932cc6d2007-06-21 13:10:21 +02001161 * that process.
1162 *
Jens Axboec66ab6f2007-06-12 21:17:17 +02001163 */
1164ssize_t splice_direct_to_actor(struct file *in, struct splice_desc *sd,
1165 splice_direct_actor *actor)
Jens Axboeb92ce552006-04-11 13:52:07 +02001166{
1167 struct pipe_inode_info *pipe;
1168 long ret, bytes;
1169 umode_t i_mode;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001170 size_t len;
Christophe Leroy0ff28d92015-05-06 17:26:47 +02001171 int i, flags, more;
Jens Axboeb92ce552006-04-11 13:52:07 +02001172
1173 /*
1174 * We require the input being a regular file, as we don't want to
1175 * randomly drop data for eg socket -> socket splicing. Use the
1176 * piped splicing for that!
1177 */
Al Viro496ad9a2013-01-23 17:07:38 -05001178 i_mode = file_inode(in)->i_mode;
Jens Axboeb92ce552006-04-11 13:52:07 +02001179 if (unlikely(!S_ISREG(i_mode) && !S_ISBLK(i_mode)))
1180 return -EINVAL;
1181
1182 /*
1183 * neither in nor out is a pipe, setup an internal pipe attached to
1184 * 'out' and transfer the wanted data from 'in' to 'out' through that
1185 */
1186 pipe = current->splice_pipe;
Jens Axboe49570e92006-04-11 13:56:09 +02001187 if (unlikely(!pipe)) {
Al Viro7bee1302013-03-21 11:04:15 -04001188 pipe = alloc_pipe_info();
Jens Axboeb92ce552006-04-11 13:52:07 +02001189 if (!pipe)
1190 return -ENOMEM;
1191
1192 /*
1193 * We don't have an immediate reader, but we'll read the stuff
Jens Axboe00522fb2006-04-26 14:39:29 +02001194 * out of the pipe right after the splice_to_pipe(). So set
Jens Axboeb92ce552006-04-11 13:52:07 +02001195 * PIPE_READERS appropriately.
1196 */
1197 pipe->readers = 1;
1198
1199 current->splice_pipe = pipe;
1200 }
1201
1202 /*
Ingo Molnar73d62d82006-04-11 13:57:21 +02001203 * Do the splice.
Jens Axboeb92ce552006-04-11 13:52:07 +02001204 */
1205 ret = 0;
1206 bytes = 0;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001207 len = sd->total_len;
1208 flags = sd->flags;
1209
1210 /*
1211 * Don't block on output, we have to drain the direct pipe.
1212 */
1213 sd->flags &= ~SPLICE_F_NONBLOCK;
Christophe Leroy0ff28d92015-05-06 17:26:47 +02001214 more = sd->flags & SPLICE_F_MORE;
Jens Axboeb92ce552006-04-11 13:52:07 +02001215
1216 while (len) {
Jens Axboe51a92c02007-07-13 14:11:43 +02001217 size_t read_len;
Tom Zanussia82c53a2008-05-09 13:28:36 +02001218 loff_t pos = sd->pos, prev_pos = pos;
Jens Axboeb92ce552006-04-11 13:52:07 +02001219
Jens Axboebcd4f3a2007-07-16 14:41:49 +02001220 ret = do_splice_to(in, &pos, pipe, len, flags);
Jens Axboe51a92c02007-07-13 14:11:43 +02001221 if (unlikely(ret <= 0))
Jens Axboeb92ce552006-04-11 13:52:07 +02001222 goto out_release;
1223
1224 read_len = ret;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001225 sd->total_len = read_len;
Jens Axboeb92ce552006-04-11 13:52:07 +02001226
1227 /*
Christophe Leroy0ff28d92015-05-06 17:26:47 +02001228 * If more data is pending, set SPLICE_F_MORE
1229 * If this is the last data and SPLICE_F_MORE was not set
1230 * initially, clears it.
1231 */
1232 if (read_len < len)
1233 sd->flags |= SPLICE_F_MORE;
1234 else if (!more)
1235 sd->flags &= ~SPLICE_F_MORE;
1236 /*
Jens Axboeb92ce552006-04-11 13:52:07 +02001237 * NOTE: nonblocking mode only applies to the input. We
1238 * must not do the output in nonblocking mode as then we
1239 * could get stuck data in the internal pipe:
1240 */
Jens Axboec66ab6f2007-06-12 21:17:17 +02001241 ret = actor(pipe, sd);
Tom Zanussia82c53a2008-05-09 13:28:36 +02001242 if (unlikely(ret <= 0)) {
1243 sd->pos = prev_pos;
Jens Axboeb92ce552006-04-11 13:52:07 +02001244 goto out_release;
Tom Zanussia82c53a2008-05-09 13:28:36 +02001245 }
Jens Axboeb92ce552006-04-11 13:52:07 +02001246
1247 bytes += ret;
1248 len -= ret;
Jens Axboebcd4f3a2007-07-16 14:41:49 +02001249 sd->pos = pos;
Jens Axboeb92ce552006-04-11 13:52:07 +02001250
Tom Zanussia82c53a2008-05-09 13:28:36 +02001251 if (ret < read_len) {
1252 sd->pos = prev_pos + ret;
Jens Axboe51a92c02007-07-13 14:11:43 +02001253 goto out_release;
Tom Zanussia82c53a2008-05-09 13:28:36 +02001254 }
Jens Axboeb92ce552006-04-11 13:52:07 +02001255 }
1256
Jens Axboe9e971982008-01-29 21:05:57 +01001257done:
Jens Axboeb92ce552006-04-11 13:52:07 +02001258 pipe->nrbufs = pipe->curbuf = 0;
Jens Axboe80848702008-01-30 12:24:48 +01001259 file_accessed(in);
Jens Axboeb92ce552006-04-11 13:52:07 +02001260 return bytes;
1261
1262out_release:
1263 /*
1264 * If we did an incomplete transfer we must release
1265 * the pipe buffers in question:
1266 */
Jens Axboe35f3d142010-05-20 10:43:18 +02001267 for (i = 0; i < pipe->buffers; i++) {
Jens Axboeb92ce552006-04-11 13:52:07 +02001268 struct pipe_buffer *buf = pipe->bufs + i;
1269
1270 if (buf->ops) {
1271 buf->ops->release(pipe, buf);
1272 buf->ops = NULL;
1273 }
1274 }
Jens Axboeb92ce552006-04-11 13:52:07 +02001275
Jens Axboe9e971982008-01-29 21:05:57 +01001276 if (!bytes)
1277 bytes = ret;
Jens Axboeb92ce552006-04-11 13:52:07 +02001278
Jens Axboe9e971982008-01-29 21:05:57 +01001279 goto done;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001280}
1281EXPORT_SYMBOL(splice_direct_to_actor);
1282
1283static int direct_splice_actor(struct pipe_inode_info *pipe,
1284 struct splice_desc *sd)
1285{
Jens Axboe6a14b902007-06-14 13:08:55 +02001286 struct file *file = sd->u.file;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001287
Al Viro7995bd22013-06-20 18:58:36 +04001288 return do_splice_from(pipe, file, sd->opos, sd->total_len,
Changli Gao2cb4b052010-06-29 13:09:18 +02001289 sd->flags);
Jens Axboec66ab6f2007-06-12 21:17:17 +02001290}
1291
Jens Axboe932cc6d2007-06-21 13:10:21 +02001292/**
1293 * do_splice_direct - splices data directly between two files
1294 * @in: file to splice from
1295 * @ppos: input file offset
1296 * @out: file to splice to
Randy Dunlapacdb37c2013-06-22 19:44:08 -07001297 * @opos: output file offset
Jens Axboe932cc6d2007-06-21 13:10:21 +02001298 * @len: number of bytes to splice
1299 * @flags: splice modifier flags
1300 *
1301 * Description:
1302 * For use by do_sendfile(). splice can easily emulate sendfile, but
1303 * doing it in the application would incur an extra system call
1304 * (splice in + splice out, as compared to just sendfile()). So this helper
1305 * can splice directly through a process-private pipe.
1306 *
1307 */
Jens Axboec66ab6f2007-06-12 21:17:17 +02001308long do_splice_direct(struct file *in, loff_t *ppos, struct file *out,
Al Viro7995bd22013-06-20 18:58:36 +04001309 loff_t *opos, size_t len, unsigned int flags)
Jens Axboec66ab6f2007-06-12 21:17:17 +02001310{
1311 struct splice_desc sd = {
1312 .len = len,
1313 .total_len = len,
1314 .flags = flags,
1315 .pos = *ppos,
Jens Axboe6a14b902007-06-14 13:08:55 +02001316 .u.file = out,
Al Viro7995bd22013-06-20 18:58:36 +04001317 .opos = opos,
Jens Axboec66ab6f2007-06-12 21:17:17 +02001318 };
Jens Axboe51a92c02007-07-13 14:11:43 +02001319 long ret;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001320
Al Viro18c67cb2013-06-19 15:41:54 +04001321 if (unlikely(!(out->f_mode & FMODE_WRITE)))
1322 return -EBADF;
1323
1324 if (unlikely(out->f_flags & O_APPEND))
1325 return -EINVAL;
1326
1327 ret = rw_verify_area(WRITE, out, opos, len);
1328 if (unlikely(ret < 0))
1329 return ret;
1330
Jens Axboec66ab6f2007-06-12 21:17:17 +02001331 ret = splice_direct_to_actor(in, &sd, direct_splice_actor);
Jens Axboe51a92c02007-07-13 14:11:43 +02001332 if (ret > 0)
Tom Zanussia82c53a2008-05-09 13:28:36 +02001333 *ppos = sd.pos;
Jens Axboe51a92c02007-07-13 14:11:43 +02001334
Jens Axboec66ab6f2007-06-12 21:17:17 +02001335 return ret;
Jens Axboeb92ce552006-04-11 13:52:07 +02001336}
Miklos Szeredi1c118592014-10-24 00:14:35 +02001337EXPORT_SYMBOL(do_splice_direct);
Jens Axboeb92ce552006-04-11 13:52:07 +02001338
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001339static int splice_pipe_to_pipe(struct pipe_inode_info *ipipe,
1340 struct pipe_inode_info *opipe,
1341 size_t len, unsigned int flags);
Jens Axboeddac0d32006-11-04 12:49:32 +01001342
1343/*
Jens Axboe83f91352006-04-02 23:05:09 +02001344 * Determine where to splice to/from.
1345 */
Ingo Molnar529565d2006-04-10 15:18:58 +02001346static long do_splice(struct file *in, loff_t __user *off_in,
1347 struct file *out, loff_t __user *off_out,
1348 size_t len, unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +02001349{
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001350 struct pipe_inode_info *ipipe;
1351 struct pipe_inode_info *opipe;
Al Viro7995bd22013-06-20 18:58:36 +04001352 loff_t offset;
Jens Axboea4514eb2006-04-19 15:57:05 +02001353 long ret;
Jens Axboe5274f052006-03-30 15:15:30 +02001354
Linus Torvalds71993e62010-11-28 13:56:09 -08001355 ipipe = get_pipe_info(in);
1356 opipe = get_pipe_info(out);
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001357
1358 if (ipipe && opipe) {
1359 if (off_in || off_out)
1360 return -ESPIPE;
1361
1362 if (!(in->f_mode & FMODE_READ))
1363 return -EBADF;
1364
1365 if (!(out->f_mode & FMODE_WRITE))
1366 return -EBADF;
1367
1368 /* Splicing to self would be fun, but... */
1369 if (ipipe == opipe)
1370 return -EINVAL;
1371
1372 return splice_pipe_to_pipe(ipipe, opipe, len, flags);
1373 }
1374
1375 if (ipipe) {
Ingo Molnar529565d2006-04-10 15:18:58 +02001376 if (off_in)
1377 return -ESPIPE;
Jens Axboeb92ce552006-04-11 13:52:07 +02001378 if (off_out) {
Changli Gao19c9a492010-06-29 13:10:36 +02001379 if (!(out->f_mode & FMODE_PWRITE))
Jens Axboeb92ce552006-04-11 13:52:07 +02001380 return -EINVAL;
Jens Axboecbb7e572006-04-11 14:57:50 +02001381 if (copy_from_user(&offset, off_out, sizeof(loff_t)))
Jens Axboeb92ce552006-04-11 13:52:07 +02001382 return -EFAULT;
Al Viro7995bd22013-06-20 18:58:36 +04001383 } else {
1384 offset = out->f_pos;
1385 }
Ingo Molnar529565d2006-04-10 15:18:58 +02001386
Al Viro18c67cb2013-06-19 15:41:54 +04001387 if (unlikely(!(out->f_mode & FMODE_WRITE)))
1388 return -EBADF;
1389
1390 if (unlikely(out->f_flags & O_APPEND))
1391 return -EINVAL;
1392
1393 ret = rw_verify_area(WRITE, out, &offset, len);
1394 if (unlikely(ret < 0))
1395 return ret;
1396
Al Viro500368f2013-05-23 20:07:11 -04001397 file_start_write(out);
Al Viro7995bd22013-06-20 18:58:36 +04001398 ret = do_splice_from(ipipe, out, &offset, len, flags);
Al Viro500368f2013-05-23 20:07:11 -04001399 file_end_write(out);
Jens Axboea4514eb2006-04-19 15:57:05 +02001400
Al Viro7995bd22013-06-20 18:58:36 +04001401 if (!off_out)
1402 out->f_pos = offset;
1403 else if (copy_to_user(off_out, &offset, sizeof(loff_t)))
Jens Axboea4514eb2006-04-19 15:57:05 +02001404 ret = -EFAULT;
1405
1406 return ret;
Ingo Molnar529565d2006-04-10 15:18:58 +02001407 }
Jens Axboe5274f052006-03-30 15:15:30 +02001408
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001409 if (opipe) {
Ingo Molnar529565d2006-04-10 15:18:58 +02001410 if (off_out)
1411 return -ESPIPE;
Jens Axboeb92ce552006-04-11 13:52:07 +02001412 if (off_in) {
Changli Gao19c9a492010-06-29 13:10:36 +02001413 if (!(in->f_mode & FMODE_PREAD))
Jens Axboeb92ce552006-04-11 13:52:07 +02001414 return -EINVAL;
Jens Axboecbb7e572006-04-11 14:57:50 +02001415 if (copy_from_user(&offset, off_in, sizeof(loff_t)))
Jens Axboeb92ce552006-04-11 13:52:07 +02001416 return -EFAULT;
Al Viro7995bd22013-06-20 18:58:36 +04001417 } else {
1418 offset = in->f_pos;
1419 }
Ingo Molnar529565d2006-04-10 15:18:58 +02001420
Al Viro7995bd22013-06-20 18:58:36 +04001421 ret = do_splice_to(in, &offset, opipe, len, flags);
Jens Axboea4514eb2006-04-19 15:57:05 +02001422
Al Viro7995bd22013-06-20 18:58:36 +04001423 if (!off_in)
1424 in->f_pos = offset;
1425 else if (copy_to_user(off_in, &offset, sizeof(loff_t)))
Jens Axboea4514eb2006-04-19 15:57:05 +02001426 ret = -EFAULT;
1427
1428 return ret;
Ingo Molnar529565d2006-04-10 15:18:58 +02001429 }
Jens Axboe5274f052006-03-30 15:15:30 +02001430
1431 return -EINVAL;
1432}
1433
Jens Axboe912d35f2006-04-26 10:59:21 +02001434/*
1435 * Map an iov into an array of pages and offset/length tupples. With the
1436 * partial_page structure, we can map several non-contiguous ranges into
1437 * our ones pages[] map instead of splitting that operation into pieces.
1438 * Could easily be exported as a generic helper for other users, in which
1439 * case one would probably want to add a 'max_nr_pages' parameter as well.
1440 */
1441static int get_iovec_page_array(const struct iovec __user *iov,
1442 unsigned int nr_vecs, struct page **pages,
Eric Dumazetbd1a68b2012-04-04 11:08:51 +02001443 struct partial_page *partial, bool aligned,
Jens Axboe35f3d142010-05-20 10:43:18 +02001444 unsigned int pipe_buffers)
Jens Axboe912d35f2006-04-26 10:59:21 +02001445{
1446 int buffers = 0, error = 0;
1447
Jens Axboe912d35f2006-04-26 10:59:21 +02001448 while (nr_vecs) {
1449 unsigned long off, npages;
Linus Torvalds75723952007-10-01 13:17:28 -07001450 struct iovec entry;
Jens Axboe912d35f2006-04-26 10:59:21 +02001451 void __user *base;
1452 size_t len;
1453 int i;
1454
Linus Torvalds75723952007-10-01 13:17:28 -07001455 error = -EFAULT;
Nick Pigginbc40d732008-07-25 19:45:26 -07001456 if (copy_from_user(&entry, iov, sizeof(entry)))
Jens Axboe912d35f2006-04-26 10:59:21 +02001457 break;
Linus Torvalds75723952007-10-01 13:17:28 -07001458
1459 base = entry.iov_base;
1460 len = entry.iov_len;
Jens Axboe912d35f2006-04-26 10:59:21 +02001461
1462 /*
1463 * Sanity check this iovec. 0 read succeeds.
1464 */
Linus Torvalds75723952007-10-01 13:17:28 -07001465 error = 0;
Jens Axboe912d35f2006-04-26 10:59:21 +02001466 if (unlikely(!len))
1467 break;
1468 error = -EFAULT;
Bastian Blank712a30e2008-02-10 16:47:57 +02001469 if (!access_ok(VERIFY_READ, base, len))
Jens Axboe912d35f2006-04-26 10:59:21 +02001470 break;
1471
1472 /*
1473 * Get this base offset and number of pages, then map
1474 * in the user pages.
1475 */
1476 off = (unsigned long) base & ~PAGE_MASK;
Jens Axboe7afa6fd2006-05-01 20:02:33 +02001477
1478 /*
1479 * If asked for alignment, the offset must be zero and the
1480 * length a multiple of the PAGE_SIZE.
1481 */
1482 error = -EINVAL;
1483 if (aligned && (off || len & ~PAGE_MASK))
1484 break;
1485
Jens Axboe912d35f2006-04-26 10:59:21 +02001486 npages = (off + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
Jens Axboe35f3d142010-05-20 10:43:18 +02001487 if (npages > pipe_buffers - buffers)
1488 npages = pipe_buffers - buffers;
Jens Axboe912d35f2006-04-26 10:59:21 +02001489
Nick Pigginbc40d732008-07-25 19:45:26 -07001490 error = get_user_pages_fast((unsigned long)base, npages,
1491 0, &pages[buffers]);
Jens Axboe912d35f2006-04-26 10:59:21 +02001492
1493 if (unlikely(error <= 0))
1494 break;
1495
1496 /*
1497 * Fill this contiguous range into the partial page map.
1498 */
1499 for (i = 0; i < error; i++) {
Jens Axboe75914892006-05-02 12:57:18 +02001500 const int plen = min_t(size_t, len, PAGE_SIZE - off);
Jens Axboe912d35f2006-04-26 10:59:21 +02001501
1502 partial[buffers].offset = off;
1503 partial[buffers].len = plen;
1504
1505 off = 0;
1506 len -= plen;
1507 buffers++;
1508 }
1509
1510 /*
1511 * We didn't complete this iov, stop here since it probably
1512 * means we have to move some of this into a pipe to
1513 * be able to continue.
1514 */
1515 if (len)
1516 break;
1517
1518 /*
1519 * Don't continue if we mapped fewer pages than we asked for,
1520 * or if we mapped the max number of pages that we have
1521 * room for.
1522 */
Jens Axboe35f3d142010-05-20 10:43:18 +02001523 if (error < npages || buffers == pipe_buffers)
Jens Axboe912d35f2006-04-26 10:59:21 +02001524 break;
1525
1526 nr_vecs--;
1527 iov++;
1528 }
1529
Jens Axboe912d35f2006-04-26 10:59:21 +02001530 if (buffers)
1531 return buffers;
1532
1533 return error;
1534}
1535
Jens Axboe6a14b902007-06-14 13:08:55 +02001536static int pipe_to_user(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
1537 struct splice_desc *sd)
1538{
Al Viro6130f532014-02-03 18:19:51 -05001539 int n = copy_page_to_iter(buf->page, buf->offset, sd->len, sd->u.data);
1540 return n == sd->len ? n : -EFAULT;
Jens Axboe6a14b902007-06-14 13:08:55 +02001541}
1542
1543/*
1544 * For lack of a better implementation, implement vmsplice() to userspace
1545 * as a simple copy of the pipes pages to the user iov.
1546 */
Al Viro6130f532014-02-03 18:19:51 -05001547static long vmsplice_to_user(struct file *file, const struct iovec __user *uiov,
Jens Axboe6a14b902007-06-14 13:08:55 +02001548 unsigned long nr_segs, unsigned int flags)
1549{
1550 struct pipe_inode_info *pipe;
1551 struct splice_desc sd;
Jens Axboe6a14b902007-06-14 13:08:55 +02001552 long ret;
Al Viro6130f532014-02-03 18:19:51 -05001553 struct iovec iovstack[UIO_FASTIOV];
1554 struct iovec *iov = iovstack;
1555 struct iov_iter iter;
Jens Axboe6a14b902007-06-14 13:08:55 +02001556
Linus Torvalds71993e62010-11-28 13:56:09 -08001557 pipe = get_pipe_info(file);
Jens Axboe6a14b902007-06-14 13:08:55 +02001558 if (!pipe)
1559 return -EBADF;
1560
Al Viro345995f2015-03-21 19:17:55 -04001561 ret = import_iovec(READ, uiov, nr_segs,
1562 ARRAY_SIZE(iovstack), &iov, &iter);
1563 if (ret < 0)
1564 return ret;
Al Viro6130f532014-02-03 18:19:51 -05001565
Al Viro345995f2015-03-21 19:17:55 -04001566 sd.total_len = iov_iter_count(&iter);
Al Viro6130f532014-02-03 18:19:51 -05001567 sd.len = 0;
Al Viro6130f532014-02-03 18:19:51 -05001568 sd.flags = flags;
1569 sd.u.data = &iter;
1570 sd.pos = 0;
1571
Al Viro345995f2015-03-21 19:17:55 -04001572 if (sd.total_len) {
1573 pipe_lock(pipe);
1574 ret = __splice_from_pipe(pipe, &sd, pipe_to_user);
1575 pipe_unlock(pipe);
1576 }
Jens Axboe6a14b902007-06-14 13:08:55 +02001577
Al Viro345995f2015-03-21 19:17:55 -04001578 kfree(iov);
Jens Axboe6a14b902007-06-14 13:08:55 +02001579 return ret;
1580}
1581
Jens Axboe912d35f2006-04-26 10:59:21 +02001582/*
1583 * vmsplice splices a user address range into a pipe. It can be thought of
1584 * as splice-from-memory, where the regular splice is splice-from-file (or
1585 * to file). In both cases the output is a pipe, naturally.
Jens Axboe912d35f2006-04-26 10:59:21 +02001586 */
Jens Axboe6a14b902007-06-14 13:08:55 +02001587static long vmsplice_to_pipe(struct file *file, const struct iovec __user *iov,
1588 unsigned long nr_segs, unsigned int flags)
Jens Axboe912d35f2006-04-26 10:59:21 +02001589{
Jens Axboeddac0d32006-11-04 12:49:32 +01001590 struct pipe_inode_info *pipe;
Jens Axboe35f3d142010-05-20 10:43:18 +02001591 struct page *pages[PIPE_DEF_BUFFERS];
1592 struct partial_page partial[PIPE_DEF_BUFFERS];
Jens Axboe912d35f2006-04-26 10:59:21 +02001593 struct splice_pipe_desc spd = {
1594 .pages = pages,
1595 .partial = partial,
Eric Dumazet047fe362012-06-12 15:24:40 +02001596 .nr_pages_max = PIPE_DEF_BUFFERS,
Jens Axboe912d35f2006-04-26 10:59:21 +02001597 .flags = flags,
1598 .ops = &user_page_pipe_buf_ops,
Jens Axboebbdfc2f2007-11-06 23:29:47 -08001599 .spd_release = spd_release_page,
Jens Axboe912d35f2006-04-26 10:59:21 +02001600 };
Jens Axboe35f3d142010-05-20 10:43:18 +02001601 long ret;
Jens Axboe912d35f2006-04-26 10:59:21 +02001602
Linus Torvalds71993e62010-11-28 13:56:09 -08001603 pipe = get_pipe_info(file);
Jens Axboeddac0d32006-11-04 12:49:32 +01001604 if (!pipe)
Jens Axboe912d35f2006-04-26 10:59:21 +02001605 return -EBADF;
Jens Axboe912d35f2006-04-26 10:59:21 +02001606
Jens Axboe35f3d142010-05-20 10:43:18 +02001607 if (splice_grow_spd(pipe, &spd))
1608 return -ENOMEM;
Jens Axboe912d35f2006-04-26 10:59:21 +02001609
Jens Axboe35f3d142010-05-20 10:43:18 +02001610 spd.nr_pages = get_iovec_page_array(iov, nr_segs, spd.pages,
Eric Dumazetbd1a68b2012-04-04 11:08:51 +02001611 spd.partial, false,
Eric Dumazet047fe362012-06-12 15:24:40 +02001612 spd.nr_pages_max);
Jens Axboe35f3d142010-05-20 10:43:18 +02001613 if (spd.nr_pages <= 0)
1614 ret = spd.nr_pages;
1615 else
1616 ret = splice_to_pipe(pipe, &spd);
1617
Eric Dumazet047fe362012-06-12 15:24:40 +02001618 splice_shrink_spd(&spd);
Jens Axboe35f3d142010-05-20 10:43:18 +02001619 return ret;
Jens Axboe912d35f2006-04-26 10:59:21 +02001620}
1621
Jens Axboe6a14b902007-06-14 13:08:55 +02001622/*
1623 * Note that vmsplice only really supports true splicing _from_ user memory
1624 * to a pipe, not the other way around. Splicing from user memory is a simple
1625 * operation that can be supported without any funky alignment restrictions
1626 * or nasty vm tricks. We simply map in the user memory and fill them into
1627 * a pipe. The reverse isn't quite as easy, though. There are two possible
1628 * solutions for that:
1629 *
1630 * - memcpy() the data internally, at which point we might as well just
1631 * do a regular read() on the buffer anyway.
1632 * - Lots of nasty vm tricks, that are neither fast nor flexible (it
1633 * has restriction limitations on both ends of the pipe).
1634 *
1635 * Currently we punt and implement it as a normal copy, see pipe_to_user().
1636 *
1637 */
Heiko Carstens836f92a2009-01-14 14:14:33 +01001638SYSCALL_DEFINE4(vmsplice, int, fd, const struct iovec __user *, iov,
1639 unsigned long, nr_segs, unsigned int, flags)
Jens Axboe912d35f2006-04-26 10:59:21 +02001640{
Al Viro2903ff02012-08-28 12:52:22 -04001641 struct fd f;
Jens Axboe912d35f2006-04-26 10:59:21 +02001642 long error;
Jens Axboe912d35f2006-04-26 10:59:21 +02001643
Jens Axboe6a14b902007-06-14 13:08:55 +02001644 if (unlikely(nr_segs > UIO_MAXIOV))
1645 return -EINVAL;
1646 else if (unlikely(!nr_segs))
1647 return 0;
1648
Jens Axboe912d35f2006-04-26 10:59:21 +02001649 error = -EBADF;
Al Viro2903ff02012-08-28 12:52:22 -04001650 f = fdget(fd);
1651 if (f.file) {
1652 if (f.file->f_mode & FMODE_WRITE)
1653 error = vmsplice_to_pipe(f.file, iov, nr_segs, flags);
1654 else if (f.file->f_mode & FMODE_READ)
1655 error = vmsplice_to_user(f.file, iov, nr_segs, flags);
Jens Axboe912d35f2006-04-26 10:59:21 +02001656
Al Viro2903ff02012-08-28 12:52:22 -04001657 fdput(f);
Jens Axboe912d35f2006-04-26 10:59:21 +02001658 }
1659
1660 return error;
1661}
1662
Al Viro76b021d2013-03-02 10:19:56 -05001663#ifdef CONFIG_COMPAT
1664COMPAT_SYSCALL_DEFINE4(vmsplice, int, fd, const struct compat_iovec __user *, iov32,
1665 unsigned int, nr_segs, unsigned int, flags)
1666{
1667 unsigned i;
1668 struct iovec __user *iov;
1669 if (nr_segs > UIO_MAXIOV)
1670 return -EINVAL;
1671 iov = compat_alloc_user_space(nr_segs * sizeof(struct iovec));
1672 for (i = 0; i < nr_segs; i++) {
1673 struct compat_iovec v;
1674 if (get_user(v.iov_base, &iov32[i].iov_base) ||
1675 get_user(v.iov_len, &iov32[i].iov_len) ||
1676 put_user(compat_ptr(v.iov_base), &iov[i].iov_base) ||
1677 put_user(v.iov_len, &iov[i].iov_len))
1678 return -EFAULT;
1679 }
1680 return sys_vmsplice(fd, iov, nr_segs, flags);
1681}
1682#endif
1683
Heiko Carstens836f92a2009-01-14 14:14:33 +01001684SYSCALL_DEFINE6(splice, int, fd_in, loff_t __user *, off_in,
1685 int, fd_out, loff_t __user *, off_out,
1686 size_t, len, unsigned int, flags)
Jens Axboe5274f052006-03-30 15:15:30 +02001687{
Al Viro2903ff02012-08-28 12:52:22 -04001688 struct fd in, out;
Jens Axboe5274f052006-03-30 15:15:30 +02001689 long error;
Jens Axboe5274f052006-03-30 15:15:30 +02001690
1691 if (unlikely(!len))
1692 return 0;
1693
1694 error = -EBADF;
Al Viro2903ff02012-08-28 12:52:22 -04001695 in = fdget(fd_in);
1696 if (in.file) {
1697 if (in.file->f_mode & FMODE_READ) {
1698 out = fdget(fd_out);
1699 if (out.file) {
1700 if (out.file->f_mode & FMODE_WRITE)
1701 error = do_splice(in.file, off_in,
1702 out.file, off_out,
Ingo Molnar529565d2006-04-10 15:18:58 +02001703 len, flags);
Al Viro2903ff02012-08-28 12:52:22 -04001704 fdput(out);
Jens Axboe5274f052006-03-30 15:15:30 +02001705 }
1706 }
Al Viro2903ff02012-08-28 12:52:22 -04001707 fdput(in);
Jens Axboe5274f052006-03-30 15:15:30 +02001708 }
Jens Axboe5274f052006-03-30 15:15:30 +02001709 return error;
1710}
Jens Axboe70524492006-04-11 15:51:17 +02001711
1712/*
Jens Axboeaadd06e2006-07-10 11:00:01 +02001713 * Make sure there's data to read. Wait for input if we can, otherwise
1714 * return an appropriate error.
1715 */
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001716static int ipipe_prep(struct pipe_inode_info *pipe, unsigned int flags)
Jens Axboeaadd06e2006-07-10 11:00:01 +02001717{
1718 int ret;
1719
1720 /*
1721 * Check ->nrbufs without the inode lock first. This function
1722 * is speculative anyways, so missing one is ok.
1723 */
1724 if (pipe->nrbufs)
1725 return 0;
1726
1727 ret = 0;
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001728 pipe_lock(pipe);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001729
1730 while (!pipe->nrbufs) {
1731 if (signal_pending(current)) {
1732 ret = -ERESTARTSYS;
1733 break;
1734 }
1735 if (!pipe->writers)
1736 break;
1737 if (!pipe->waiting_writers) {
1738 if (flags & SPLICE_F_NONBLOCK) {
1739 ret = -EAGAIN;
1740 break;
1741 }
1742 }
1743 pipe_wait(pipe);
1744 }
1745
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001746 pipe_unlock(pipe);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001747 return ret;
1748}
1749
1750/*
1751 * Make sure there's writeable room. Wait for room if we can, otherwise
1752 * return an appropriate error.
1753 */
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001754static int opipe_prep(struct pipe_inode_info *pipe, unsigned int flags)
Jens Axboeaadd06e2006-07-10 11:00:01 +02001755{
1756 int ret;
1757
1758 /*
1759 * Check ->nrbufs without the inode lock first. This function
1760 * is speculative anyways, so missing one is ok.
1761 */
Jens Axboe35f3d142010-05-20 10:43:18 +02001762 if (pipe->nrbufs < pipe->buffers)
Jens Axboeaadd06e2006-07-10 11:00:01 +02001763 return 0;
1764
1765 ret = 0;
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001766 pipe_lock(pipe);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001767
Jens Axboe35f3d142010-05-20 10:43:18 +02001768 while (pipe->nrbufs >= pipe->buffers) {
Jens Axboeaadd06e2006-07-10 11:00:01 +02001769 if (!pipe->readers) {
1770 send_sig(SIGPIPE, current, 0);
1771 ret = -EPIPE;
1772 break;
1773 }
1774 if (flags & SPLICE_F_NONBLOCK) {
1775 ret = -EAGAIN;
1776 break;
1777 }
1778 if (signal_pending(current)) {
1779 ret = -ERESTARTSYS;
1780 break;
1781 }
1782 pipe->waiting_writers++;
1783 pipe_wait(pipe);
1784 pipe->waiting_writers--;
1785 }
1786
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001787 pipe_unlock(pipe);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001788 return ret;
1789}
1790
1791/*
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001792 * Splice contents of ipipe to opipe.
1793 */
1794static int splice_pipe_to_pipe(struct pipe_inode_info *ipipe,
1795 struct pipe_inode_info *opipe,
1796 size_t len, unsigned int flags)
1797{
1798 struct pipe_buffer *ibuf, *obuf;
1799 int ret = 0, nbuf;
1800 bool input_wakeup = false;
1801
1802
1803retry:
1804 ret = ipipe_prep(ipipe, flags);
1805 if (ret)
1806 return ret;
1807
1808 ret = opipe_prep(opipe, flags);
1809 if (ret)
1810 return ret;
1811
1812 /*
1813 * Potential ABBA deadlock, work around it by ordering lock
1814 * grabbing by pipe info address. Otherwise two different processes
1815 * could deadlock (one doing tee from A -> B, the other from B -> A).
1816 */
1817 pipe_double_lock(ipipe, opipe);
1818
1819 do {
1820 if (!opipe->readers) {
1821 send_sig(SIGPIPE, current, 0);
1822 if (!ret)
1823 ret = -EPIPE;
1824 break;
1825 }
1826
1827 if (!ipipe->nrbufs && !ipipe->writers)
1828 break;
1829
1830 /*
1831 * Cannot make any progress, because either the input
1832 * pipe is empty or the output pipe is full.
1833 */
Jens Axboe35f3d142010-05-20 10:43:18 +02001834 if (!ipipe->nrbufs || opipe->nrbufs >= opipe->buffers) {
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001835 /* Already processed some buffers, break */
1836 if (ret)
1837 break;
1838
1839 if (flags & SPLICE_F_NONBLOCK) {
1840 ret = -EAGAIN;
1841 break;
1842 }
1843
1844 /*
1845 * We raced with another reader/writer and haven't
1846 * managed to process any buffers. A zero return
1847 * value means EOF, so retry instead.
1848 */
1849 pipe_unlock(ipipe);
1850 pipe_unlock(opipe);
1851 goto retry;
1852 }
1853
1854 ibuf = ipipe->bufs + ipipe->curbuf;
Jens Axboe35f3d142010-05-20 10:43:18 +02001855 nbuf = (opipe->curbuf + opipe->nrbufs) & (opipe->buffers - 1);
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001856 obuf = opipe->bufs + nbuf;
1857
1858 if (len >= ibuf->len) {
1859 /*
1860 * Simply move the whole buffer from ipipe to opipe
1861 */
1862 *obuf = *ibuf;
1863 ibuf->ops = NULL;
1864 opipe->nrbufs++;
Jens Axboe35f3d142010-05-20 10:43:18 +02001865 ipipe->curbuf = (ipipe->curbuf + 1) & (ipipe->buffers - 1);
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001866 ipipe->nrbufs--;
1867 input_wakeup = true;
1868 } else {
1869 /*
1870 * Get a reference to this pipe buffer,
1871 * so we can copy the contents over.
1872 */
1873 ibuf->ops->get(ipipe, ibuf);
1874 *obuf = *ibuf;
1875
1876 /*
1877 * Don't inherit the gift flag, we need to
1878 * prevent multiple steals of this page.
1879 */
1880 obuf->flags &= ~PIPE_BUF_FLAG_GIFT;
1881
1882 obuf->len = len;
1883 opipe->nrbufs++;
1884 ibuf->offset += obuf->len;
1885 ibuf->len -= obuf->len;
1886 }
1887 ret += obuf->len;
1888 len -= obuf->len;
1889 } while (len);
1890
1891 pipe_unlock(ipipe);
1892 pipe_unlock(opipe);
1893
1894 /*
1895 * If we put data in the output pipe, wakeup any potential readers.
1896 */
Namhyung Kim825cdcb2011-05-23 19:58:53 +02001897 if (ret > 0)
1898 wakeup_pipe_readers(opipe);
1899
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001900 if (input_wakeup)
1901 wakeup_pipe_writers(ipipe);
1902
1903 return ret;
1904}
1905
1906/*
Jens Axboe70524492006-04-11 15:51:17 +02001907 * Link contents of ipipe to opipe.
1908 */
1909static int link_pipe(struct pipe_inode_info *ipipe,
1910 struct pipe_inode_info *opipe,
1911 size_t len, unsigned int flags)
1912{
1913 struct pipe_buffer *ibuf, *obuf;
Jens Axboeaadd06e2006-07-10 11:00:01 +02001914 int ret = 0, i = 0, nbuf;
Jens Axboe70524492006-04-11 15:51:17 +02001915
1916 /*
1917 * Potential ABBA deadlock, work around it by ordering lock
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001918 * grabbing by pipe info address. Otherwise two different processes
Jens Axboe70524492006-04-11 15:51:17 +02001919 * could deadlock (one doing tee from A -> B, the other from B -> A).
1920 */
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001921 pipe_double_lock(ipipe, opipe);
Jens Axboe70524492006-04-11 15:51:17 +02001922
Jens Axboeaadd06e2006-07-10 11:00:01 +02001923 do {
Jens Axboe70524492006-04-11 15:51:17 +02001924 if (!opipe->readers) {
1925 send_sig(SIGPIPE, current, 0);
1926 if (!ret)
1927 ret = -EPIPE;
1928 break;
1929 }
Jens Axboe70524492006-04-11 15:51:17 +02001930
1931 /*
Jens Axboeaadd06e2006-07-10 11:00:01 +02001932 * If we have iterated all input buffers or ran out of
1933 * output room, break.
Jens Axboe70524492006-04-11 15:51:17 +02001934 */
Jens Axboe35f3d142010-05-20 10:43:18 +02001935 if (i >= ipipe->nrbufs || opipe->nrbufs >= opipe->buffers)
Jens Axboe70524492006-04-11 15:51:17 +02001936 break;
Jens Axboeaadd06e2006-07-10 11:00:01 +02001937
Jens Axboe35f3d142010-05-20 10:43:18 +02001938 ibuf = ipipe->bufs + ((ipipe->curbuf + i) & (ipipe->buffers-1));
1939 nbuf = (opipe->curbuf + opipe->nrbufs) & (opipe->buffers - 1);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001940
Jens Axboe2a27250e2006-04-19 15:56:40 +02001941 /*
Jens Axboeaadd06e2006-07-10 11:00:01 +02001942 * Get a reference to this pipe buffer,
1943 * so we can copy the contents over.
Jens Axboe2a27250e2006-04-19 15:56:40 +02001944 */
Jens Axboeaadd06e2006-07-10 11:00:01 +02001945 ibuf->ops->get(ipipe, ibuf);
Jens Axboe70524492006-04-11 15:51:17 +02001946
Jens Axboeaadd06e2006-07-10 11:00:01 +02001947 obuf = opipe->bufs + nbuf;
1948 *obuf = *ibuf;
Jens Axboe70524492006-04-11 15:51:17 +02001949
Jens Axboeaadd06e2006-07-10 11:00:01 +02001950 /*
1951 * Don't inherit the gift flag, we need to
1952 * prevent multiple steals of this page.
1953 */
1954 obuf->flags &= ~PIPE_BUF_FLAG_GIFT;
1955
1956 if (obuf->len > len)
1957 obuf->len = len;
1958
1959 opipe->nrbufs++;
1960 ret += obuf->len;
1961 len -= obuf->len;
1962 i++;
1963 } while (len);
Jens Axboe70524492006-04-11 15:51:17 +02001964
Jens Axboe02cf01a2008-02-20 10:34:51 +01001965 /*
1966 * return EAGAIN if we have the potential of some data in the
1967 * future, otherwise just return 0
1968 */
1969 if (!ret && ipipe->waiting_writers && (flags & SPLICE_F_NONBLOCK))
1970 ret = -EAGAIN;
1971
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001972 pipe_unlock(ipipe);
1973 pipe_unlock(opipe);
Jens Axboe70524492006-04-11 15:51:17 +02001974
Jens Axboeaadd06e2006-07-10 11:00:01 +02001975 /*
1976 * If we put data in the output pipe, wakeup any potential readers.
1977 */
Namhyung Kim825cdcb2011-05-23 19:58:53 +02001978 if (ret > 0)
1979 wakeup_pipe_readers(opipe);
Jens Axboe70524492006-04-11 15:51:17 +02001980
1981 return ret;
1982}
1983
1984/*
1985 * This is a tee(1) implementation that works on pipes. It doesn't copy
1986 * any data, it simply references the 'in' pages on the 'out' pipe.
1987 * The 'flags' used are the SPLICE_F_* variants, currently the only
1988 * applicable one is SPLICE_F_NONBLOCK.
1989 */
1990static long do_tee(struct file *in, struct file *out, size_t len,
1991 unsigned int flags)
1992{
Linus Torvalds71993e62010-11-28 13:56:09 -08001993 struct pipe_inode_info *ipipe = get_pipe_info(in);
1994 struct pipe_inode_info *opipe = get_pipe_info(out);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001995 int ret = -EINVAL;
Jens Axboe70524492006-04-11 15:51:17 +02001996
1997 /*
Jens Axboeaadd06e2006-07-10 11:00:01 +02001998 * Duplicate the contents of ipipe to opipe without actually
1999 * copying the data.
Jens Axboe70524492006-04-11 15:51:17 +02002000 */
Jens Axboeaadd06e2006-07-10 11:00:01 +02002001 if (ipipe && opipe && ipipe != opipe) {
2002 /*
2003 * Keep going, unless we encounter an error. The ipipe/opipe
2004 * ordering doesn't really matter.
2005 */
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02002006 ret = ipipe_prep(ipipe, flags);
Jens Axboeaadd06e2006-07-10 11:00:01 +02002007 if (!ret) {
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02002008 ret = opipe_prep(opipe, flags);
Jens Axboe02cf01a2008-02-20 10:34:51 +01002009 if (!ret)
Jens Axboeaadd06e2006-07-10 11:00:01 +02002010 ret = link_pipe(ipipe, opipe, len, flags);
Jens Axboeaadd06e2006-07-10 11:00:01 +02002011 }
2012 }
Jens Axboe70524492006-04-11 15:51:17 +02002013
Jens Axboeaadd06e2006-07-10 11:00:01 +02002014 return ret;
Jens Axboe70524492006-04-11 15:51:17 +02002015}
2016
Heiko Carstens836f92a2009-01-14 14:14:33 +01002017SYSCALL_DEFINE4(tee, int, fdin, int, fdout, size_t, len, unsigned int, flags)
Jens Axboe70524492006-04-11 15:51:17 +02002018{
Al Viro2903ff02012-08-28 12:52:22 -04002019 struct fd in;
2020 int error;
Jens Axboe70524492006-04-11 15:51:17 +02002021
2022 if (unlikely(!len))
2023 return 0;
2024
2025 error = -EBADF;
Al Viro2903ff02012-08-28 12:52:22 -04002026 in = fdget(fdin);
2027 if (in.file) {
2028 if (in.file->f_mode & FMODE_READ) {
2029 struct fd out = fdget(fdout);
2030 if (out.file) {
2031 if (out.file->f_mode & FMODE_WRITE)
2032 error = do_tee(in.file, out.file,
2033 len, flags);
2034 fdput(out);
Jens Axboe70524492006-04-11 15:51:17 +02002035 }
2036 }
Al Viro2903ff02012-08-28 12:52:22 -04002037 fdput(in);
Jens Axboe70524492006-04-11 15:51:17 +02002038 }
2039
2040 return error;
2041}