blob: bfe62ae40f40920e6b95fa8ce16cc3130b8b0972 [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}
264
Hugh Dickins708e3502011-07-25 17:12:32 -0700265void spd_release_page(struct splice_pipe_desc *spd, unsigned int i)
Jens Axboebbdfc2f2007-11-06 23:29:47 -0800266{
267 page_cache_release(spd->pages[i]);
268}
269
Jens Axboe35f3d142010-05-20 10:43:18 +0200270/*
271 * Check if we need to grow the arrays holding pages and partial page
272 * descriptions.
273 */
Eric Dumazet047fe362012-06-12 15:24:40 +0200274int splice_grow_spd(const struct pipe_inode_info *pipe, struct splice_pipe_desc *spd)
Jens Axboe35f3d142010-05-20 10:43:18 +0200275{
Eric Dumazet047fe362012-06-12 15:24:40 +0200276 unsigned int buffers = ACCESS_ONCE(pipe->buffers);
277
278 spd->nr_pages_max = buffers;
279 if (buffers <= PIPE_DEF_BUFFERS)
Jens Axboe35f3d142010-05-20 10:43:18 +0200280 return 0;
281
Eric Dumazet047fe362012-06-12 15:24:40 +0200282 spd->pages = kmalloc(buffers * sizeof(struct page *), GFP_KERNEL);
283 spd->partial = kmalloc(buffers * sizeof(struct partial_page), GFP_KERNEL);
Jens Axboe35f3d142010-05-20 10:43:18 +0200284
285 if (spd->pages && spd->partial)
286 return 0;
287
288 kfree(spd->pages);
289 kfree(spd->partial);
290 return -ENOMEM;
291}
292
Eric Dumazet047fe362012-06-12 15:24:40 +0200293void splice_shrink_spd(struct splice_pipe_desc *spd)
Jens Axboe35f3d142010-05-20 10:43:18 +0200294{
Eric Dumazet047fe362012-06-12 15:24:40 +0200295 if (spd->nr_pages_max <= PIPE_DEF_BUFFERS)
Jens Axboe35f3d142010-05-20 10:43:18 +0200296 return;
297
298 kfree(spd->pages);
299 kfree(spd->partial);
300}
301
Ingo Molnar3a326a22006-04-10 15:18:35 +0200302static int
Jens Axboecbb7e572006-04-11 14:57:50 +0200303__generic_file_splice_read(struct file *in, loff_t *ppos,
304 struct pipe_inode_info *pipe, size_t len,
305 unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +0200306{
307 struct address_space *mapping = in->f_mapping;
Fengguang Wud8983912007-07-19 01:48:06 -0700308 unsigned int loff, nr_pages, req_pages;
Jens Axboe35f3d142010-05-20 10:43:18 +0200309 struct page *pages[PIPE_DEF_BUFFERS];
310 struct partial_page partial[PIPE_DEF_BUFFERS];
Jens Axboe5274f052006-03-30 15:15:30 +0200311 struct page *page;
Jens Axboe91ad66e2006-04-19 15:55:10 +0200312 pgoff_t index, end_index;
313 loff_t isize;
Jens Axboeeb207962006-04-27 11:05:22 +0200314 int error, page_nr;
Jens Axboe912d35f2006-04-26 10:59:21 +0200315 struct splice_pipe_desc spd = {
316 .pages = pages,
317 .partial = partial,
Eric Dumazet047fe362012-06-12 15:24:40 +0200318 .nr_pages_max = PIPE_DEF_BUFFERS,
Jens Axboe912d35f2006-04-26 10:59:21 +0200319 .flags = flags,
320 .ops = &page_cache_pipe_buf_ops,
Jens Axboebbdfc2f2007-11-06 23:29:47 -0800321 .spd_release = spd_release_page,
Jens Axboe912d35f2006-04-26 10:59:21 +0200322 };
Jens Axboe5274f052006-03-30 15:15:30 +0200323
Jens Axboe35f3d142010-05-20 10:43:18 +0200324 if (splice_grow_spd(pipe, &spd))
325 return -ENOMEM;
326
Jens Axboecbb7e572006-04-11 14:57:50 +0200327 index = *ppos >> PAGE_CACHE_SHIFT;
Jens Axboe912d35f2006-04-26 10:59:21 +0200328 loff = *ppos & ~PAGE_CACHE_MASK;
Fengguang Wud8983912007-07-19 01:48:06 -0700329 req_pages = (len + loff + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
Eric Dumazet047fe362012-06-12 15:24:40 +0200330 nr_pages = min(req_pages, spd.nr_pages_max);
Jens Axboe5274f052006-03-30 15:15:30 +0200331
332 /*
Jens Axboeeb207962006-04-27 11:05:22 +0200333 * Lookup the (hopefully) full range of pages we need.
334 */
Jens Axboe35f3d142010-05-20 10:43:18 +0200335 spd.nr_pages = find_get_pages_contig(mapping, index, nr_pages, spd.pages);
Fengguang Wu431a4822007-07-19 01:48:05 -0700336 index += spd.nr_pages;
Jens Axboeeb207962006-04-27 11:05:22 +0200337
338 /*
339 * If find_get_pages_contig() returned fewer pages than we needed,
Fengguang Wu431a4822007-07-19 01:48:05 -0700340 * readahead/allocate the rest and fill in the holes.
Jens Axboeeb207962006-04-27 11:05:22 +0200341 */
Fengguang Wu431a4822007-07-19 01:48:05 -0700342 if (spd.nr_pages < nr_pages)
Rusty Russellcf914a72007-07-19 01:48:08 -0700343 page_cache_sync_readahead(mapping, &in->f_ra, in,
344 index, req_pages - spd.nr_pages);
Fengguang Wu431a4822007-07-19 01:48:05 -0700345
Jens Axboe932cc6d2007-06-21 13:10:21 +0200346 error = 0;
Jens Axboeeb207962006-04-27 11:05:22 +0200347 while (spd.nr_pages < nr_pages) {
348 /*
349 * Page could be there, find_get_pages_contig() breaks on
350 * the first hole.
351 */
352 page = find_get_page(mapping, index);
353 if (!page) {
Jens Axboee27dedd2006-05-01 19:59:54 +0200354 /*
Jens Axboeeb207962006-04-27 11:05:22 +0200355 * page didn't exist, allocate one.
356 */
357 page = page_cache_alloc_cold(mapping);
358 if (!page)
359 break;
360
361 error = add_to_page_cache_lru(page, mapping, index,
Nick Piggin0ae0b5d2010-05-25 10:25:26 +0200362 GFP_KERNEL);
Jens Axboeeb207962006-04-27 11:05:22 +0200363 if (unlikely(error)) {
364 page_cache_release(page);
Jens Axboea0548872006-05-03 10:58:22 +0200365 if (error == -EEXIST)
366 continue;
Jens Axboeeb207962006-04-27 11:05:22 +0200367 break;
368 }
369 /*
370 * add_to_page_cache() locks the page, unlock it
371 * to avoid convoluting the logic below even more.
372 */
373 unlock_page(page);
374 }
375
Jens Axboe35f3d142010-05-20 10:43:18 +0200376 spd.pages[spd.nr_pages++] = page;
Jens Axboeeb207962006-04-27 11:05:22 +0200377 index++;
378 }
379
380 /*
381 * Now loop over the map and see if we need to start IO on any
382 * pages, fill in the partial map, etc.
383 */
384 index = *ppos >> PAGE_CACHE_SHIFT;
385 nr_pages = spd.nr_pages;
386 spd.nr_pages = 0;
387 for (page_nr = 0; page_nr < nr_pages; page_nr++) {
Jens Axboe82aa5d62006-04-20 13:05:48 +0200388 unsigned int this_len;
389
390 if (!len)
391 break;
392
393 /*
394 * this_len is the max we'll use from this page
395 */
Andrew Mortonba5f5d92006-04-25 15:33:34 +0200396 this_len = min_t(unsigned long, len, PAGE_CACHE_SIZE - loff);
Jens Axboe35f3d142010-05-20 10:43:18 +0200397 page = spd.pages[page_nr];
Jens Axboe7480a902006-04-11 13:52:47 +0200398
Fengguang Wua08a1662007-07-19 01:48:03 -0700399 if (PageReadahead(page))
Rusty Russellcf914a72007-07-19 01:48:08 -0700400 page_cache_async_readahead(mapping, &in->f_ra, in,
Fengguang Wud8983912007-07-19 01:48:06 -0700401 page, index, req_pages - page_nr);
Fengguang Wua08a1662007-07-19 01:48:03 -0700402
Jens Axboe7480a902006-04-11 13:52:47 +0200403 /*
404 * If the page isn't uptodate, we may need to start io on it
405 */
406 if (!PageUptodate(page)) {
Miklos Szeredi69650312010-08-03 12:48:50 +0200407 lock_page(page);
Jens Axboe7480a902006-04-11 13:52:47 +0200408
409 /*
Miklos Szeredi32502b82008-07-04 09:35:17 +0200410 * Page was truncated, or invalidated by the
411 * filesystem. Redo the find/create, but this time the
412 * page is kept locked, so there's no chance of another
413 * race with truncate/invalidate.
Jens Axboe7480a902006-04-11 13:52:47 +0200414 */
415 if (!page->mapping) {
416 unlock_page(page);
Miklos Szeredi32502b82008-07-04 09:35:17 +0200417 page = find_or_create_page(mapping, index,
418 mapping_gfp_mask(mapping));
419
420 if (!page) {
421 error = -ENOMEM;
422 break;
423 }
Jens Axboe35f3d142010-05-20 10:43:18 +0200424 page_cache_release(spd.pages[page_nr]);
425 spd.pages[page_nr] = page;
Jens Axboe7480a902006-04-11 13:52:47 +0200426 }
427 /*
428 * page was already under io and is now done, great
429 */
430 if (PageUptodate(page)) {
431 unlock_page(page);
432 goto fill_it;
433 }
434
Jens Axboe7480a902006-04-11 13:52:47 +0200435 /*
436 * need to read in the page
437 */
438 error = mapping->a_ops->readpage(in, page);
Jens Axboe7480a902006-04-11 13:52:47 +0200439 if (unlikely(error)) {
Jens Axboeeb207962006-04-27 11:05:22 +0200440 /*
441 * We really should re-lookup the page here,
442 * but it complicates things a lot. Instead
443 * lets just do what we already stored, and
444 * we'll get it the next time we are called.
445 */
Jens Axboe7480a902006-04-11 13:52:47 +0200446 if (error == AOP_TRUNCATED_PAGE)
Jens Axboeeb207962006-04-27 11:05:22 +0200447 error = 0;
448
Jens Axboe7480a902006-04-11 13:52:47 +0200449 break;
450 }
Jens Axboe620a3242007-06-07 09:39:42 +0200451 }
452fill_it:
453 /*
454 * i_size must be checked after PageUptodate.
455 */
456 isize = i_size_read(mapping->host);
457 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
458 if (unlikely(!isize || index > end_index))
459 break;
460
461 /*
462 * if this is the last page, see if we need to shrink
463 * the length and stop
464 */
465 if (end_index == index) {
466 unsigned int plen;
Jens Axboe91ad66e2006-04-19 15:55:10 +0200467
468 /*
Jens Axboe620a3242007-06-07 09:39:42 +0200469 * max good bytes in this page
Jens Axboe91ad66e2006-04-19 15:55:10 +0200470 */
Jens Axboe620a3242007-06-07 09:39:42 +0200471 plen = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
472 if (plen <= loff)
Jens Axboe91ad66e2006-04-19 15:55:10 +0200473 break;
Jens Axboe91ad66e2006-04-19 15:55:10 +0200474
475 /*
Jens Axboe620a3242007-06-07 09:39:42 +0200476 * force quit after adding this page
Jens Axboe91ad66e2006-04-19 15:55:10 +0200477 */
Jens Axboe620a3242007-06-07 09:39:42 +0200478 this_len = min(this_len, plen - loff);
479 len = this_len;
Jens Axboe7480a902006-04-11 13:52:47 +0200480 }
Jens Axboe620a3242007-06-07 09:39:42 +0200481
Jens Axboe35f3d142010-05-20 10:43:18 +0200482 spd.partial[page_nr].offset = loff;
483 spd.partial[page_nr].len = this_len;
Jens Axboe82aa5d62006-04-20 13:05:48 +0200484 len -= this_len;
Jens Axboe91ad66e2006-04-19 15:55:10 +0200485 loff = 0;
Jens Axboeeb207962006-04-27 11:05:22 +0200486 spd.nr_pages++;
487 index++;
Jens Axboe5274f052006-03-30 15:15:30 +0200488 }
489
Jens Axboeeb207962006-04-27 11:05:22 +0200490 /*
Hugh Dickins475ecad2007-06-07 09:36:00 +0200491 * Release any pages at the end, if we quit early. 'page_nr' is how far
Jens Axboeeb207962006-04-27 11:05:22 +0200492 * we got, 'nr_pages' is how many pages are in the map.
493 */
494 while (page_nr < nr_pages)
Jens Axboe35f3d142010-05-20 10:43:18 +0200495 page_cache_release(spd.pages[page_nr++]);
Fengguang Wuf4e6b492007-10-16 01:24:33 -0700496 in->f_ra.prev_pos = (loff_t)index << PAGE_CACHE_SHIFT;
Jens Axboeeb207962006-04-27 11:05:22 +0200497
Jens Axboe912d35f2006-04-26 10:59:21 +0200498 if (spd.nr_pages)
Jens Axboe35f3d142010-05-20 10:43:18 +0200499 error = splice_to_pipe(pipe, &spd);
Jens Axboe5274f052006-03-30 15:15:30 +0200500
Eric Dumazet047fe362012-06-12 15:24:40 +0200501 splice_shrink_spd(&spd);
Jens Axboe7480a902006-04-11 13:52:47 +0200502 return error;
Jens Axboe5274f052006-03-30 15:15:30 +0200503}
504
Jens Axboe83f91352006-04-02 23:05:09 +0200505/**
506 * generic_file_splice_read - splice data from file to a pipe
507 * @in: file to splice from
Jens Axboe932cc6d2007-06-21 13:10:21 +0200508 * @ppos: position in @in
Jens Axboe83f91352006-04-02 23:05:09 +0200509 * @pipe: pipe to splice to
510 * @len: number of bytes to splice
511 * @flags: splice modifier flags
512 *
Jens Axboe932cc6d2007-06-21 13:10:21 +0200513 * Description:
514 * Will read pages from given file and fill them into a pipe. Can be
515 * used as long as the address_space operations for the source implements
516 * a readpage() hook.
517 *
Jens Axboe83f91352006-04-02 23:05:09 +0200518 */
Jens Axboecbb7e572006-04-11 14:57:50 +0200519ssize_t generic_file_splice_read(struct file *in, loff_t *ppos,
520 struct pipe_inode_info *pipe, size_t len,
521 unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +0200522{
Jens Axboed366d3982007-06-01 14:54:11 +0200523 loff_t isize, left;
Jens Axboe8191ecd2008-04-10 08:24:25 +0200524 int ret;
Jens Axboed366d3982007-06-01 14:54:11 +0200525
Boaz Harroshbe64f882015-04-15 16:15:17 -0700526 if (IS_DAX(in->f_mapping->host))
527 return default_file_splice_read(in, ppos, pipe, len, flags);
528
Jens Axboed366d3982007-06-01 14:54:11 +0200529 isize = i_size_read(in->f_mapping->host);
530 if (unlikely(*ppos >= isize))
531 return 0;
532
533 left = isize - *ppos;
534 if (unlikely(left < len))
535 len = left;
Jens Axboe5274f052006-03-30 15:15:30 +0200536
Jens Axboe8191ecd2008-04-10 08:24:25 +0200537 ret = __generic_file_splice_read(in, ppos, pipe, len, flags);
Miklos Szeredi723590e2009-08-15 08:43:22 +0200538 if (ret > 0) {
Jens Axboecbb7e572006-04-11 14:57:50 +0200539 *ppos += ret;
Miklos Szeredi723590e2009-08-15 08:43:22 +0200540 file_accessed(in);
541 }
Jens Axboe5274f052006-03-30 15:15:30 +0200542
543 return ret;
544}
Jens Axboe059a8f32006-04-02 23:06:05 +0200545EXPORT_SYMBOL(generic_file_splice_read);
546
Miklos Szeredi68181732009-05-07 15:37:36 +0200547static const struct pipe_buf_operations default_pipe_buf_ops = {
548 .can_merge = 0,
Miklos Szeredi68181732009-05-07 15:37:36 +0200549 .confirm = generic_pipe_buf_confirm,
550 .release = generic_pipe_buf_release,
551 .steal = generic_pipe_buf_steal,
552 .get = generic_pipe_buf_get,
553};
554
Miklos Szeredi28a625c2014-01-22 19:36:57 +0100555static int generic_pipe_buf_nosteal(struct pipe_inode_info *pipe,
556 struct pipe_buffer *buf)
557{
558 return 1;
559}
560
561/* Pipe buffer operations for a socket and similar. */
562const struct pipe_buf_operations nosteal_pipe_buf_ops = {
563 .can_merge = 0,
Miklos Szeredi28a625c2014-01-22 19:36:57 +0100564 .confirm = generic_pipe_buf_confirm,
565 .release = generic_pipe_buf_release,
566 .steal = generic_pipe_buf_nosteal,
567 .get = generic_pipe_buf_get,
568};
569EXPORT_SYMBOL(nosteal_pipe_buf_ops);
570
Miklos Szeredi68181732009-05-07 15:37:36 +0200571static ssize_t kernel_readv(struct file *file, const struct iovec *vec,
572 unsigned long vlen, loff_t offset)
573{
574 mm_segment_t old_fs;
575 loff_t pos = offset;
576 ssize_t res;
577
578 old_fs = get_fs();
579 set_fs(get_ds());
580 /* The cast to a user pointer is valid due to the set_fs() */
581 res = vfs_readv(file, (const struct iovec __user *)vec, vlen, &pos);
582 set_fs(old_fs);
583
584 return res;
585}
586
Al Viro7bb307e2013-02-23 14:51:48 -0500587ssize_t kernel_write(struct file *file, const char *buf, size_t count,
Miklos Szeredib2858d72009-05-19 11:37:46 +0200588 loff_t pos)
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +0200589{
590 mm_segment_t old_fs;
591 ssize_t res;
592
593 old_fs = get_fs();
594 set_fs(get_ds());
595 /* The cast to a user pointer is valid due to the set_fs() */
Al Viro7bb307e2013-02-23 14:51:48 -0500596 res = vfs_write(file, (__force const char __user *)buf, count, &pos);
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +0200597 set_fs(old_fs);
598
599 return res;
600}
Al Viro7bb307e2013-02-23 14:51:48 -0500601EXPORT_SYMBOL(kernel_write);
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +0200602
Miklos Szeredi68181732009-05-07 15:37:36 +0200603ssize_t default_file_splice_read(struct file *in, loff_t *ppos,
604 struct pipe_inode_info *pipe, size_t len,
605 unsigned int flags)
606{
607 unsigned int nr_pages;
608 unsigned int nr_freed;
609 size_t offset;
Jens Axboe35f3d142010-05-20 10:43:18 +0200610 struct page *pages[PIPE_DEF_BUFFERS];
611 struct partial_page partial[PIPE_DEF_BUFFERS];
612 struct iovec *vec, __vec[PIPE_DEF_BUFFERS];
Miklos Szeredi68181732009-05-07 15:37:36 +0200613 ssize_t res;
614 size_t this_len;
615 int error;
616 int i;
617 struct splice_pipe_desc spd = {
618 .pages = pages,
619 .partial = partial,
Eric Dumazet047fe362012-06-12 15:24:40 +0200620 .nr_pages_max = PIPE_DEF_BUFFERS,
Miklos Szeredi68181732009-05-07 15:37:36 +0200621 .flags = flags,
622 .ops = &default_pipe_buf_ops,
623 .spd_release = spd_release_page,
624 };
625
Jens Axboe35f3d142010-05-20 10:43:18 +0200626 if (splice_grow_spd(pipe, &spd))
627 return -ENOMEM;
628
629 res = -ENOMEM;
630 vec = __vec;
Eric Dumazet047fe362012-06-12 15:24:40 +0200631 if (spd.nr_pages_max > PIPE_DEF_BUFFERS) {
632 vec = kmalloc(spd.nr_pages_max * sizeof(struct iovec), GFP_KERNEL);
Jens Axboe35f3d142010-05-20 10:43:18 +0200633 if (!vec)
634 goto shrink_ret;
635 }
636
Miklos Szeredi68181732009-05-07 15:37:36 +0200637 offset = *ppos & ~PAGE_CACHE_MASK;
638 nr_pages = (len + offset + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
639
Eric Dumazet047fe362012-06-12 15:24:40 +0200640 for (i = 0; i < nr_pages && i < spd.nr_pages_max && len; i++) {
Miklos Szeredi68181732009-05-07 15:37:36 +0200641 struct page *page;
642
Jens Axboe4f231222009-05-13 08:35:35 +0200643 page = alloc_page(GFP_USER);
Miklos Szeredi68181732009-05-07 15:37:36 +0200644 error = -ENOMEM;
645 if (!page)
646 goto err;
647
648 this_len = min_t(size_t, len, PAGE_CACHE_SIZE - offset);
Jens Axboe4f231222009-05-13 08:35:35 +0200649 vec[i].iov_base = (void __user *) page_address(page);
Miklos Szeredi68181732009-05-07 15:37:36 +0200650 vec[i].iov_len = this_len;
Jens Axboe35f3d142010-05-20 10:43:18 +0200651 spd.pages[i] = page;
Miklos Szeredi68181732009-05-07 15:37:36 +0200652 spd.nr_pages++;
653 len -= this_len;
654 offset = 0;
655 }
656
657 res = kernel_readv(in, vec, spd.nr_pages, *ppos);
Andrew Morton77f6bf52009-05-14 09:49:44 +0200658 if (res < 0) {
659 error = res;
Miklos Szeredi68181732009-05-07 15:37:36 +0200660 goto err;
Andrew Morton77f6bf52009-05-14 09:49:44 +0200661 }
Miklos Szeredi68181732009-05-07 15:37:36 +0200662
663 error = 0;
664 if (!res)
665 goto err;
666
667 nr_freed = 0;
668 for (i = 0; i < spd.nr_pages; i++) {
Miklos Szeredi68181732009-05-07 15:37:36 +0200669 this_len = min_t(size_t, vec[i].iov_len, res);
Jens Axboe35f3d142010-05-20 10:43:18 +0200670 spd.partial[i].offset = 0;
671 spd.partial[i].len = this_len;
Miklos Szeredi68181732009-05-07 15:37:36 +0200672 if (!this_len) {
Jens Axboe35f3d142010-05-20 10:43:18 +0200673 __free_page(spd.pages[i]);
674 spd.pages[i] = NULL;
Miklos Szeredi68181732009-05-07 15:37:36 +0200675 nr_freed++;
676 }
677 res -= this_len;
678 }
679 spd.nr_pages -= nr_freed;
680
681 res = splice_to_pipe(pipe, &spd);
682 if (res > 0)
683 *ppos += res;
684
Jens Axboe35f3d142010-05-20 10:43:18 +0200685shrink_ret:
686 if (vec != __vec)
687 kfree(vec);
Eric Dumazet047fe362012-06-12 15:24:40 +0200688 splice_shrink_spd(&spd);
Miklos Szeredi68181732009-05-07 15:37:36 +0200689 return res;
690
691err:
Jens Axboe4f231222009-05-13 08:35:35 +0200692 for (i = 0; i < spd.nr_pages; i++)
Jens Axboe35f3d142010-05-20 10:43:18 +0200693 __free_page(spd.pages[i]);
Jens Axboe4f231222009-05-13 08:35:35 +0200694
Jens Axboe35f3d142010-05-20 10:43:18 +0200695 res = error;
696 goto shrink_ret;
Miklos Szeredi68181732009-05-07 15:37:36 +0200697}
698EXPORT_SYMBOL(default_file_splice_read);
699
Jens Axboe5274f052006-03-30 15:15:30 +0200700/*
Jens Axboe4f6f0bd2006-04-02 23:04:46 +0200701 * Send 'sd->len' bytes to socket from 'sd->file' at position 'sd->pos'
Jens Axboe016b6612006-04-25 15:42:00 +0200702 * using sendpage(). Return the number of bytes sent.
Jens Axboe5274f052006-03-30 15:15:30 +0200703 */
Jens Axboe76ad4d12006-05-03 10:41:33 +0200704static int pipe_to_sendpage(struct pipe_inode_info *pipe,
Jens Axboe5274f052006-03-30 15:15:30 +0200705 struct pipe_buffer *buf, struct splice_desc *sd)
706{
Jens Axboe6a14b902007-06-14 13:08:55 +0200707 struct file *file = sd->u.file;
Jens Axboe5274f052006-03-30 15:15:30 +0200708 loff_t pos = sd->pos;
Michał Mirosława8adbe32010-12-17 08:56:44 +0100709 int more;
Jens Axboe5274f052006-03-30 15:15:30 +0200710
Al Viro72c2d532013-09-22 16:27:52 -0400711 if (!likely(file->f_op->sendpage))
Michał Mirosława8adbe32010-12-17 08:56:44 +0100712 return -EINVAL;
Jens Axboe5274f052006-03-30 15:15:30 +0200713
Eric Dumazet35f9c092012-04-05 03:05:35 +0000714 more = (sd->flags & SPLICE_F_MORE) ? MSG_MORE : 0;
Eric Dumazetae62ca72013-01-06 18:21:49 +0000715
716 if (sd->len < sd->total_len && pipe->nrbufs > 1)
Eric Dumazet35f9c092012-04-05 03:05:35 +0000717 more |= MSG_SENDPAGE_NOTLAST;
Eric Dumazetae62ca72013-01-06 18:21:49 +0000718
Michał Mirosława8adbe32010-12-17 08:56:44 +0100719 return file->f_op->sendpage(file, buf->page, buf->offset,
720 sd->len, &pos, more);
Jens Axboe5274f052006-03-30 15:15:30 +0200721}
722
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200723static void wakeup_pipe_writers(struct pipe_inode_info *pipe)
724{
725 smp_mb();
726 if (waitqueue_active(&pipe->wait))
727 wake_up_interruptible(&pipe->wait);
728 kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
729}
730
731/**
732 * splice_from_pipe_feed - feed available data from a pipe to a file
733 * @pipe: pipe to splice from
734 * @sd: information to @actor
735 * @actor: handler that splices the data
736 *
737 * Description:
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200738 * This function loops over the pipe and calls @actor to do the
739 * actual moving of a single struct pipe_buffer to the desired
740 * destination. It returns when there's no more buffers left in
741 * the pipe or if the requested number of bytes (@sd->total_len)
742 * have been copied. It returns a positive number (one) if the
743 * pipe needs to be filled with more data, zero if the required
744 * number of bytes have been copied and -errno on error.
745 *
746 * This, together with splice_from_pipe_{begin,end,next}, may be
747 * used to implement the functionality of __splice_from_pipe() when
748 * locking is required around copying the pipe buffers to the
749 * destination.
750 */
Al Viro96f9bc82014-04-05 04:35:49 -0400751static int splice_from_pipe_feed(struct pipe_inode_info *pipe, struct splice_desc *sd,
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200752 splice_actor *actor)
753{
754 int ret;
755
756 while (pipe->nrbufs) {
757 struct pipe_buffer *buf = pipe->bufs + pipe->curbuf;
758 const struct pipe_buf_operations *ops = buf->ops;
759
760 sd->len = buf->len;
761 if (sd->len > sd->total_len)
762 sd->len = sd->total_len;
763
Michał Mirosława8adbe32010-12-17 08:56:44 +0100764 ret = buf->ops->confirm(pipe, buf);
765 if (unlikely(ret)) {
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200766 if (ret == -ENODATA)
767 ret = 0;
768 return ret;
769 }
Michał Mirosława8adbe32010-12-17 08:56:44 +0100770
771 ret = actor(pipe, buf, sd);
772 if (ret <= 0)
773 return ret;
774
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200775 buf->offset += ret;
776 buf->len -= ret;
777
778 sd->num_spliced += ret;
779 sd->len -= ret;
780 sd->pos += ret;
781 sd->total_len -= ret;
782
783 if (!buf->len) {
784 buf->ops = NULL;
785 ops->release(pipe, buf);
Jens Axboe35f3d142010-05-20 10:43:18 +0200786 pipe->curbuf = (pipe->curbuf + 1) & (pipe->buffers - 1);
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200787 pipe->nrbufs--;
Al Viro6447a3c2013-03-21 11:01:38 -0400788 if (pipe->files)
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200789 sd->need_wakeup = true;
790 }
791
792 if (!sd->total_len)
793 return 0;
794 }
795
796 return 1;
797}
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200798
799/**
800 * splice_from_pipe_next - wait for some data to splice from
801 * @pipe: pipe to splice from
802 * @sd: information about the splice operation
803 *
804 * Description:
805 * This function will wait for some data and return a positive
806 * value (one) if pipe buffers are available. It will return zero
807 * or -errno if no more data needs to be spliced.
808 */
Al Viro96f9bc82014-04-05 04:35:49 -0400809static int splice_from_pipe_next(struct pipe_inode_info *pipe, struct splice_desc *sd)
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200810{
811 while (!pipe->nrbufs) {
812 if (!pipe->writers)
813 return 0;
814
815 if (!pipe->waiting_writers && sd->num_spliced)
816 return 0;
817
818 if (sd->flags & SPLICE_F_NONBLOCK)
819 return -EAGAIN;
820
821 if (signal_pending(current))
822 return -ERESTARTSYS;
823
824 if (sd->need_wakeup) {
825 wakeup_pipe_writers(pipe);
826 sd->need_wakeup = false;
827 }
828
829 pipe_wait(pipe);
830 }
831
832 return 1;
833}
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200834
835/**
836 * splice_from_pipe_begin - start splicing from pipe
Randy Dunlapb80901b2009-04-16 19:09:55 -0700837 * @sd: information about the splice operation
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200838 *
839 * Description:
840 * This function should be called before a loop containing
841 * splice_from_pipe_next() and splice_from_pipe_feed() to
842 * initialize the necessary fields of @sd.
843 */
Al Viro96f9bc82014-04-05 04:35:49 -0400844static void splice_from_pipe_begin(struct splice_desc *sd)
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200845{
846 sd->num_spliced = 0;
847 sd->need_wakeup = false;
848}
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200849
850/**
851 * splice_from_pipe_end - finish splicing from pipe
852 * @pipe: pipe to splice from
853 * @sd: information about the splice operation
854 *
855 * Description:
856 * This function will wake up pipe writers if necessary. It should
857 * be called after a loop containing splice_from_pipe_next() and
858 * splice_from_pipe_feed().
859 */
Al Viro96f9bc82014-04-05 04:35:49 -0400860static void splice_from_pipe_end(struct pipe_inode_info *pipe, struct splice_desc *sd)
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200861{
862 if (sd->need_wakeup)
863 wakeup_pipe_writers(pipe);
864}
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200865
Jens Axboe932cc6d2007-06-21 13:10:21 +0200866/**
867 * __splice_from_pipe - splice data from a pipe to given actor
868 * @pipe: pipe to splice from
869 * @sd: information to @actor
870 * @actor: handler that splices the data
871 *
872 * Description:
873 * This function does little more than loop over the pipe and call
874 * @actor to do the actual moving of a single struct pipe_buffer to
875 * the desired destination. See pipe_to_file, pipe_to_sendpage, or
876 * pipe_to_user.
877 *
Jens Axboe83f91352006-04-02 23:05:09 +0200878 */
Jens Axboec66ab6f2007-06-12 21:17:17 +0200879ssize_t __splice_from_pipe(struct pipe_inode_info *pipe, struct splice_desc *sd,
880 splice_actor *actor)
Jens Axboe5274f052006-03-30 15:15:30 +0200881{
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200882 int ret;
Jens Axboe5274f052006-03-30 15:15:30 +0200883
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200884 splice_from_pipe_begin(sd);
885 do {
886 ret = splice_from_pipe_next(pipe, sd);
887 if (ret > 0)
888 ret = splice_from_pipe_feed(pipe, sd, actor);
889 } while (ret > 0);
890 splice_from_pipe_end(pipe, sd);
Jens Axboe5274f052006-03-30 15:15:30 +0200891
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200892 return sd->num_spliced ? sd->num_spliced : ret;
Jens Axboe5274f052006-03-30 15:15:30 +0200893}
Mark Fasheh40bee44e2007-03-21 13:11:02 +0100894EXPORT_SYMBOL(__splice_from_pipe);
Jens Axboe5274f052006-03-30 15:15:30 +0200895
Jens Axboe932cc6d2007-06-21 13:10:21 +0200896/**
897 * splice_from_pipe - splice data from a pipe to a file
898 * @pipe: pipe to splice from
899 * @out: file to splice to
900 * @ppos: position in @out
901 * @len: how many bytes to splice
902 * @flags: splice modifier flags
903 * @actor: handler that splices the data
904 *
905 * Description:
Miklos Szeredi29339702009-04-14 19:48:37 +0200906 * See __splice_from_pipe. This function locks the pipe inode,
Jens Axboe932cc6d2007-06-21 13:10:21 +0200907 * otherwise it's identical to __splice_from_pipe().
908 *
909 */
Mark Fasheh6da61802006-10-17 18:43:07 +0200910ssize_t splice_from_pipe(struct pipe_inode_info *pipe, struct file *out,
911 loff_t *ppos, size_t len, unsigned int flags,
912 splice_actor *actor)
913{
914 ssize_t ret;
Jens Axboec66ab6f2007-06-12 21:17:17 +0200915 struct splice_desc sd = {
916 .total_len = len,
917 .flags = flags,
918 .pos = *ppos,
Jens Axboe6a14b902007-06-14 13:08:55 +0200919 .u.file = out,
Jens Axboec66ab6f2007-06-12 21:17:17 +0200920 };
Mark Fasheh6da61802006-10-17 18:43:07 +0200921
Miklos Szeredi61e0d472009-04-14 19:48:41 +0200922 pipe_lock(pipe);
Jens Axboec66ab6f2007-06-12 21:17:17 +0200923 ret = __splice_from_pipe(pipe, &sd, actor);
Miklos Szeredi61e0d472009-04-14 19:48:41 +0200924 pipe_unlock(pipe);
Mark Fasheh6da61802006-10-17 18:43:07 +0200925
926 return ret;
927}
928
929/**
Al Viro8d020762014-04-05 04:27:08 -0400930 * iter_file_splice_write - splice data from a pipe to a file
931 * @pipe: pipe info
932 * @out: file to write to
933 * @ppos: position in @out
934 * @len: number of bytes to splice
935 * @flags: splice modifier flags
936 *
937 * Description:
938 * Will either move or copy pages (determined by @flags options) from
939 * the given pipe inode to the given file.
940 * This one is ->write_iter-based.
941 *
942 */
943ssize_t
944iter_file_splice_write(struct pipe_inode_info *pipe, struct file *out,
945 loff_t *ppos, size_t len, unsigned int flags)
946{
947 struct splice_desc sd = {
948 .total_len = len,
949 .flags = flags,
950 .pos = *ppos,
951 .u.file = out,
952 };
953 int nbufs = pipe->buffers;
954 struct bio_vec *array = kcalloc(nbufs, sizeof(struct bio_vec),
955 GFP_KERNEL);
956 ssize_t ret;
957
958 if (unlikely(!array))
959 return -ENOMEM;
960
961 pipe_lock(pipe);
962
963 splice_from_pipe_begin(&sd);
964 while (sd.total_len) {
965 struct iov_iter from;
Al Viro8d020762014-04-05 04:27:08 -0400966 size_t left;
967 int n, idx;
968
969 ret = splice_from_pipe_next(pipe, &sd);
970 if (ret <= 0)
971 break;
972
973 if (unlikely(nbufs < pipe->buffers)) {
974 kfree(array);
975 nbufs = pipe->buffers;
976 array = kcalloc(nbufs, sizeof(struct bio_vec),
977 GFP_KERNEL);
978 if (!array) {
979 ret = -ENOMEM;
980 break;
981 }
982 }
983
984 /* build the vector */
985 left = sd.total_len;
986 for (n = 0, idx = pipe->curbuf; left && n < pipe->nrbufs; n++, idx++) {
987 struct pipe_buffer *buf = pipe->bufs + idx;
988 size_t this_len = buf->len;
989
990 if (this_len > left)
991 this_len = left;
992
993 if (idx == pipe->buffers - 1)
994 idx = -1;
995
996 ret = buf->ops->confirm(pipe, buf);
997 if (unlikely(ret)) {
998 if (ret == -ENODATA)
999 ret = 0;
1000 goto done;
1001 }
1002
1003 array[n].bv_page = buf->page;
1004 array[n].bv_len = this_len;
1005 array[n].bv_offset = buf->offset;
1006 left -= this_len;
1007 }
1008
Al Viro05afcb72015-01-23 01:08:07 -05001009 iov_iter_bvec(&from, ITER_BVEC | WRITE, array, n,
1010 sd.total_len - left);
Christoph Hellwigdbe4e192015-01-25 21:11:59 +01001011 ret = vfs_iter_write(out, &from, &sd.pos);
Al Viro8d020762014-04-05 04:27:08 -04001012 if (ret <= 0)
1013 break;
1014
1015 sd.num_spliced += ret;
1016 sd.total_len -= ret;
Christoph Hellwigdbe4e192015-01-25 21:11:59 +01001017 *ppos = sd.pos;
Al Viro8d020762014-04-05 04:27:08 -04001018
1019 /* dismiss the fully eaten buffers, adjust the partial one */
1020 while (ret) {
1021 struct pipe_buffer *buf = pipe->bufs + pipe->curbuf;
1022 if (ret >= buf->len) {
1023 const struct pipe_buf_operations *ops = buf->ops;
1024 ret -= buf->len;
1025 buf->len = 0;
1026 buf->ops = NULL;
1027 ops->release(pipe, buf);
1028 pipe->curbuf = (pipe->curbuf + 1) & (pipe->buffers - 1);
1029 pipe->nrbufs--;
1030 if (pipe->files)
1031 sd.need_wakeup = true;
1032 } else {
1033 buf->offset += ret;
1034 buf->len -= ret;
1035 ret = 0;
1036 }
1037 }
1038 }
1039done:
1040 kfree(array);
1041 splice_from_pipe_end(pipe, &sd);
1042
1043 pipe_unlock(pipe);
1044
1045 if (sd.num_spliced)
1046 ret = sd.num_spliced;
1047
1048 return ret;
1049}
1050
1051EXPORT_SYMBOL(iter_file_splice_write);
1052
Miklos Szeredib2858d72009-05-19 11:37:46 +02001053static int write_pipe_buf(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
1054 struct splice_desc *sd)
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +02001055{
Miklos Szeredib2858d72009-05-19 11:37:46 +02001056 int ret;
1057 void *data;
Al Viro06ae43f2013-03-20 13:19:30 -04001058 loff_t tmp = sd->pos;
Miklos Szeredib2858d72009-05-19 11:37:46 +02001059
Al Virofbb32752014-02-02 21:09:54 -05001060 data = kmap(buf->page);
Al Viro06ae43f2013-03-20 13:19:30 -04001061 ret = __kernel_write(sd->u.file, data + buf->offset, sd->len, &tmp);
Al Virofbb32752014-02-02 21:09:54 -05001062 kunmap(buf->page);
Miklos Szeredib2858d72009-05-19 11:37:46 +02001063
1064 return ret;
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +02001065}
1066
1067static ssize_t default_file_splice_write(struct pipe_inode_info *pipe,
1068 struct file *out, loff_t *ppos,
1069 size_t len, unsigned int flags)
1070{
Miklos Szeredib2858d72009-05-19 11:37:46 +02001071 ssize_t ret;
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +02001072
Miklos Szeredib2858d72009-05-19 11:37:46 +02001073 ret = splice_from_pipe(pipe, out, ppos, len, flags, write_pipe_buf);
1074 if (ret > 0)
1075 *ppos += ret;
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +02001076
Miklos Szeredib2858d72009-05-19 11:37:46 +02001077 return ret;
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +02001078}
1079
Jens Axboe83f91352006-04-02 23:05:09 +02001080/**
1081 * generic_splice_sendpage - splice data from a pipe to a socket
Jens Axboe932cc6d2007-06-21 13:10:21 +02001082 * @pipe: pipe to splice from
Jens Axboe83f91352006-04-02 23:05:09 +02001083 * @out: socket to write to
Jens Axboe932cc6d2007-06-21 13:10:21 +02001084 * @ppos: position in @out
Jens Axboe83f91352006-04-02 23:05:09 +02001085 * @len: number of bytes to splice
1086 * @flags: splice modifier flags
1087 *
Jens Axboe932cc6d2007-06-21 13:10:21 +02001088 * Description:
1089 * Will send @len bytes from the pipe to a network socket. No data copying
1090 * is involved.
Jens Axboe83f91352006-04-02 23:05:09 +02001091 *
1092 */
Ingo Molnar3a326a22006-04-10 15:18:35 +02001093ssize_t generic_splice_sendpage(struct pipe_inode_info *pipe, struct file *out,
Jens Axboecbb7e572006-04-11 14:57:50 +02001094 loff_t *ppos, size_t len, unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +02001095{
Jens Axboe00522fb2006-04-26 14:39:29 +02001096 return splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_sendpage);
Jens Axboe5274f052006-03-30 15:15:30 +02001097}
1098
Jens Axboe059a8f32006-04-02 23:06:05 +02001099EXPORT_SYMBOL(generic_splice_sendpage);
Jeff Garzika0f06782006-03-30 23:06:13 -05001100
Jens Axboe83f91352006-04-02 23:05:09 +02001101/*
1102 * Attempt to initiate a splice from pipe to file.
1103 */
Ingo Molnar3a326a22006-04-10 15:18:35 +02001104static long do_splice_from(struct pipe_inode_info *pipe, struct file *out,
Jens Axboecbb7e572006-04-11 14:57:50 +02001105 loff_t *ppos, size_t len, unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +02001106{
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +02001107 ssize_t (*splice_write)(struct pipe_inode_info *, struct file *,
1108 loff_t *, size_t, unsigned int);
Jens Axboe5274f052006-03-30 15:15:30 +02001109
Al Viro72c2d532013-09-22 16:27:52 -04001110 if (out->f_op->splice_write)
Changli Gaocc56f7d2009-11-04 09:09:52 +01001111 splice_write = out->f_op->splice_write;
1112 else
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +02001113 splice_write = default_file_splice_write;
1114
Al Viro500368f2013-05-23 20:07:11 -04001115 return splice_write(pipe, out, ppos, len, flags);
Jens Axboe5274f052006-03-30 15:15:30 +02001116}
1117
Jens Axboe83f91352006-04-02 23:05:09 +02001118/*
1119 * Attempt to initiate a splice from a file to a pipe.
1120 */
Jens Axboecbb7e572006-04-11 14:57:50 +02001121static long do_splice_to(struct file *in, loff_t *ppos,
1122 struct pipe_inode_info *pipe, size_t len,
1123 unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +02001124{
Miklos Szeredi68181732009-05-07 15:37:36 +02001125 ssize_t (*splice_read)(struct file *, loff_t *,
1126 struct pipe_inode_info *, size_t, unsigned int);
Jens Axboe5274f052006-03-30 15:15:30 +02001127 int ret;
1128
Jens Axboe49570e92006-04-11 13:56:09 +02001129 if (unlikely(!(in->f_mode & FMODE_READ)))
Jens Axboe5274f052006-03-30 15:15:30 +02001130 return -EBADF;
1131
Jens Axboecbb7e572006-04-11 14:57:50 +02001132 ret = rw_verify_area(READ, in, ppos, len);
Jens Axboe5274f052006-03-30 15:15:30 +02001133 if (unlikely(ret < 0))
1134 return ret;
1135
Al Viro72c2d532013-09-22 16:27:52 -04001136 if (in->f_op->splice_read)
Changli Gaocc56f7d2009-11-04 09:09:52 +01001137 splice_read = in->f_op->splice_read;
1138 else
Miklos Szeredi68181732009-05-07 15:37:36 +02001139 splice_read = default_file_splice_read;
1140
1141 return splice_read(in, ppos, pipe, len, flags);
Jens Axboe5274f052006-03-30 15:15:30 +02001142}
1143
Jens Axboe932cc6d2007-06-21 13:10:21 +02001144/**
1145 * splice_direct_to_actor - splices data directly between two non-pipes
1146 * @in: file to splice from
1147 * @sd: actor information on where to splice to
1148 * @actor: handles the data splicing
1149 *
1150 * Description:
1151 * This is a special case helper to splice directly between two
1152 * points, without requiring an explicit pipe. Internally an allocated
Randy Dunlap79685b82007-07-27 08:08:51 +02001153 * pipe is cached in the process, and reused during the lifetime of
Jens Axboe932cc6d2007-06-21 13:10:21 +02001154 * that process.
1155 *
Jens Axboec66ab6f2007-06-12 21:17:17 +02001156 */
1157ssize_t splice_direct_to_actor(struct file *in, struct splice_desc *sd,
1158 splice_direct_actor *actor)
Jens Axboeb92ce552006-04-11 13:52:07 +02001159{
1160 struct pipe_inode_info *pipe;
1161 long ret, bytes;
1162 umode_t i_mode;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001163 size_t len;
Christophe Leroy0ff28d92015-05-06 17:26:47 +02001164 int i, flags, more;
Jens Axboeb92ce552006-04-11 13:52:07 +02001165
1166 /*
1167 * We require the input being a regular file, as we don't want to
1168 * randomly drop data for eg socket -> socket splicing. Use the
1169 * piped splicing for that!
1170 */
Al Viro496ad9a2013-01-23 17:07:38 -05001171 i_mode = file_inode(in)->i_mode;
Jens Axboeb92ce552006-04-11 13:52:07 +02001172 if (unlikely(!S_ISREG(i_mode) && !S_ISBLK(i_mode)))
1173 return -EINVAL;
1174
1175 /*
1176 * neither in nor out is a pipe, setup an internal pipe attached to
1177 * 'out' and transfer the wanted data from 'in' to 'out' through that
1178 */
1179 pipe = current->splice_pipe;
Jens Axboe49570e92006-04-11 13:56:09 +02001180 if (unlikely(!pipe)) {
Al Viro7bee1302013-03-21 11:04:15 -04001181 pipe = alloc_pipe_info();
Jens Axboeb92ce552006-04-11 13:52:07 +02001182 if (!pipe)
1183 return -ENOMEM;
1184
1185 /*
1186 * We don't have an immediate reader, but we'll read the stuff
Jens Axboe00522fb2006-04-26 14:39:29 +02001187 * out of the pipe right after the splice_to_pipe(). So set
Jens Axboeb92ce552006-04-11 13:52:07 +02001188 * PIPE_READERS appropriately.
1189 */
1190 pipe->readers = 1;
1191
1192 current->splice_pipe = pipe;
1193 }
1194
1195 /*
Ingo Molnar73d62d82006-04-11 13:57:21 +02001196 * Do the splice.
Jens Axboeb92ce552006-04-11 13:52:07 +02001197 */
1198 ret = 0;
1199 bytes = 0;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001200 len = sd->total_len;
1201 flags = sd->flags;
1202
1203 /*
1204 * Don't block on output, we have to drain the direct pipe.
1205 */
1206 sd->flags &= ~SPLICE_F_NONBLOCK;
Christophe Leroy0ff28d92015-05-06 17:26:47 +02001207 more = sd->flags & SPLICE_F_MORE;
Jens Axboeb92ce552006-04-11 13:52:07 +02001208
1209 while (len) {
Jens Axboe51a92c02007-07-13 14:11:43 +02001210 size_t read_len;
Tom Zanussia82c53a2008-05-09 13:28:36 +02001211 loff_t pos = sd->pos, prev_pos = pos;
Jens Axboeb92ce552006-04-11 13:52:07 +02001212
Jens Axboebcd4f3a2007-07-16 14:41:49 +02001213 ret = do_splice_to(in, &pos, pipe, len, flags);
Jens Axboe51a92c02007-07-13 14:11:43 +02001214 if (unlikely(ret <= 0))
Jens Axboeb92ce552006-04-11 13:52:07 +02001215 goto out_release;
1216
1217 read_len = ret;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001218 sd->total_len = read_len;
Jens Axboeb92ce552006-04-11 13:52:07 +02001219
1220 /*
Christophe Leroy0ff28d92015-05-06 17:26:47 +02001221 * If more data is pending, set SPLICE_F_MORE
1222 * If this is the last data and SPLICE_F_MORE was not set
1223 * initially, clears it.
1224 */
1225 if (read_len < len)
1226 sd->flags |= SPLICE_F_MORE;
1227 else if (!more)
1228 sd->flags &= ~SPLICE_F_MORE;
1229 /*
Jens Axboeb92ce552006-04-11 13:52:07 +02001230 * NOTE: nonblocking mode only applies to the input. We
1231 * must not do the output in nonblocking mode as then we
1232 * could get stuck data in the internal pipe:
1233 */
Jens Axboec66ab6f2007-06-12 21:17:17 +02001234 ret = actor(pipe, sd);
Tom Zanussia82c53a2008-05-09 13:28:36 +02001235 if (unlikely(ret <= 0)) {
1236 sd->pos = prev_pos;
Jens Axboeb92ce552006-04-11 13:52:07 +02001237 goto out_release;
Tom Zanussia82c53a2008-05-09 13:28:36 +02001238 }
Jens Axboeb92ce552006-04-11 13:52:07 +02001239
1240 bytes += ret;
1241 len -= ret;
Jens Axboebcd4f3a2007-07-16 14:41:49 +02001242 sd->pos = pos;
Jens Axboeb92ce552006-04-11 13:52:07 +02001243
Tom Zanussia82c53a2008-05-09 13:28:36 +02001244 if (ret < read_len) {
1245 sd->pos = prev_pos + ret;
Jens Axboe51a92c02007-07-13 14:11:43 +02001246 goto out_release;
Tom Zanussia82c53a2008-05-09 13:28:36 +02001247 }
Jens Axboeb92ce552006-04-11 13:52:07 +02001248 }
1249
Jens Axboe9e971982008-01-29 21:05:57 +01001250done:
Jens Axboeb92ce552006-04-11 13:52:07 +02001251 pipe->nrbufs = pipe->curbuf = 0;
Jens Axboe80848702008-01-30 12:24:48 +01001252 file_accessed(in);
Jens Axboeb92ce552006-04-11 13:52:07 +02001253 return bytes;
1254
1255out_release:
1256 /*
1257 * If we did an incomplete transfer we must release
1258 * the pipe buffers in question:
1259 */
Jens Axboe35f3d142010-05-20 10:43:18 +02001260 for (i = 0; i < pipe->buffers; i++) {
Jens Axboeb92ce552006-04-11 13:52:07 +02001261 struct pipe_buffer *buf = pipe->bufs + i;
1262
1263 if (buf->ops) {
1264 buf->ops->release(pipe, buf);
1265 buf->ops = NULL;
1266 }
1267 }
Jens Axboeb92ce552006-04-11 13:52:07 +02001268
Jens Axboe9e971982008-01-29 21:05:57 +01001269 if (!bytes)
1270 bytes = ret;
Jens Axboeb92ce552006-04-11 13:52:07 +02001271
Jens Axboe9e971982008-01-29 21:05:57 +01001272 goto done;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001273}
1274EXPORT_SYMBOL(splice_direct_to_actor);
1275
1276static int direct_splice_actor(struct pipe_inode_info *pipe,
1277 struct splice_desc *sd)
1278{
Jens Axboe6a14b902007-06-14 13:08:55 +02001279 struct file *file = sd->u.file;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001280
Al Viro7995bd22013-06-20 18:58:36 +04001281 return do_splice_from(pipe, file, sd->opos, sd->total_len,
Changli Gao2cb4b052010-06-29 13:09:18 +02001282 sd->flags);
Jens Axboec66ab6f2007-06-12 21:17:17 +02001283}
1284
Jens Axboe932cc6d2007-06-21 13:10:21 +02001285/**
1286 * do_splice_direct - splices data directly between two files
1287 * @in: file to splice from
1288 * @ppos: input file offset
1289 * @out: file to splice to
Randy Dunlapacdb37c2013-06-22 19:44:08 -07001290 * @opos: output file offset
Jens Axboe932cc6d2007-06-21 13:10:21 +02001291 * @len: number of bytes to splice
1292 * @flags: splice modifier flags
1293 *
1294 * Description:
1295 * For use by do_sendfile(). splice can easily emulate sendfile, but
1296 * doing it in the application would incur an extra system call
1297 * (splice in + splice out, as compared to just sendfile()). So this helper
1298 * can splice directly through a process-private pipe.
1299 *
1300 */
Jens Axboec66ab6f2007-06-12 21:17:17 +02001301long do_splice_direct(struct file *in, loff_t *ppos, struct file *out,
Al Viro7995bd22013-06-20 18:58:36 +04001302 loff_t *opos, size_t len, unsigned int flags)
Jens Axboec66ab6f2007-06-12 21:17:17 +02001303{
1304 struct splice_desc sd = {
1305 .len = len,
1306 .total_len = len,
1307 .flags = flags,
1308 .pos = *ppos,
Jens Axboe6a14b902007-06-14 13:08:55 +02001309 .u.file = out,
Al Viro7995bd22013-06-20 18:58:36 +04001310 .opos = opos,
Jens Axboec66ab6f2007-06-12 21:17:17 +02001311 };
Jens Axboe51a92c02007-07-13 14:11:43 +02001312 long ret;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001313
Al Viro18c67cb2013-06-19 15:41:54 +04001314 if (unlikely(!(out->f_mode & FMODE_WRITE)))
1315 return -EBADF;
1316
1317 if (unlikely(out->f_flags & O_APPEND))
1318 return -EINVAL;
1319
1320 ret = rw_verify_area(WRITE, out, opos, len);
1321 if (unlikely(ret < 0))
1322 return ret;
1323
Jens Axboec66ab6f2007-06-12 21:17:17 +02001324 ret = splice_direct_to_actor(in, &sd, direct_splice_actor);
Jens Axboe51a92c02007-07-13 14:11:43 +02001325 if (ret > 0)
Tom Zanussia82c53a2008-05-09 13:28:36 +02001326 *ppos = sd.pos;
Jens Axboe51a92c02007-07-13 14:11:43 +02001327
Jens Axboec66ab6f2007-06-12 21:17:17 +02001328 return ret;
Jens Axboeb92ce552006-04-11 13:52:07 +02001329}
Miklos Szeredi1c118592014-10-24 00:14:35 +02001330EXPORT_SYMBOL(do_splice_direct);
Jens Axboeb92ce552006-04-11 13:52:07 +02001331
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001332static int splice_pipe_to_pipe(struct pipe_inode_info *ipipe,
1333 struct pipe_inode_info *opipe,
1334 size_t len, unsigned int flags);
Jens Axboeddac0d32006-11-04 12:49:32 +01001335
1336/*
Jens Axboe83f91352006-04-02 23:05:09 +02001337 * Determine where to splice to/from.
1338 */
Ingo Molnar529565d2006-04-10 15:18:58 +02001339static long do_splice(struct file *in, loff_t __user *off_in,
1340 struct file *out, loff_t __user *off_out,
1341 size_t len, unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +02001342{
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001343 struct pipe_inode_info *ipipe;
1344 struct pipe_inode_info *opipe;
Al Viro7995bd22013-06-20 18:58:36 +04001345 loff_t offset;
Jens Axboea4514eb2006-04-19 15:57:05 +02001346 long ret;
Jens Axboe5274f052006-03-30 15:15:30 +02001347
Linus Torvalds71993e62010-11-28 13:56:09 -08001348 ipipe = get_pipe_info(in);
1349 opipe = get_pipe_info(out);
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001350
1351 if (ipipe && opipe) {
1352 if (off_in || off_out)
1353 return -ESPIPE;
1354
1355 if (!(in->f_mode & FMODE_READ))
1356 return -EBADF;
1357
1358 if (!(out->f_mode & FMODE_WRITE))
1359 return -EBADF;
1360
1361 /* Splicing to self would be fun, but... */
1362 if (ipipe == opipe)
1363 return -EINVAL;
1364
1365 return splice_pipe_to_pipe(ipipe, opipe, len, flags);
1366 }
1367
1368 if (ipipe) {
Ingo Molnar529565d2006-04-10 15:18:58 +02001369 if (off_in)
1370 return -ESPIPE;
Jens Axboeb92ce552006-04-11 13:52:07 +02001371 if (off_out) {
Changli Gao19c9a492010-06-29 13:10:36 +02001372 if (!(out->f_mode & FMODE_PWRITE))
Jens Axboeb92ce552006-04-11 13:52:07 +02001373 return -EINVAL;
Jens Axboecbb7e572006-04-11 14:57:50 +02001374 if (copy_from_user(&offset, off_out, sizeof(loff_t)))
Jens Axboeb92ce552006-04-11 13:52:07 +02001375 return -EFAULT;
Al Viro7995bd22013-06-20 18:58:36 +04001376 } else {
1377 offset = out->f_pos;
1378 }
Ingo Molnar529565d2006-04-10 15:18:58 +02001379
Al Viro18c67cb2013-06-19 15:41:54 +04001380 if (unlikely(!(out->f_mode & FMODE_WRITE)))
1381 return -EBADF;
1382
1383 if (unlikely(out->f_flags & O_APPEND))
1384 return -EINVAL;
1385
1386 ret = rw_verify_area(WRITE, out, &offset, len);
1387 if (unlikely(ret < 0))
1388 return ret;
1389
Al Viro500368f2013-05-23 20:07:11 -04001390 file_start_write(out);
Al Viro7995bd22013-06-20 18:58:36 +04001391 ret = do_splice_from(ipipe, out, &offset, len, flags);
Al Viro500368f2013-05-23 20:07:11 -04001392 file_end_write(out);
Jens Axboea4514eb2006-04-19 15:57:05 +02001393
Al Viro7995bd22013-06-20 18:58:36 +04001394 if (!off_out)
1395 out->f_pos = offset;
1396 else if (copy_to_user(off_out, &offset, sizeof(loff_t)))
Jens Axboea4514eb2006-04-19 15:57:05 +02001397 ret = -EFAULT;
1398
1399 return ret;
Ingo Molnar529565d2006-04-10 15:18:58 +02001400 }
Jens Axboe5274f052006-03-30 15:15:30 +02001401
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001402 if (opipe) {
Ingo Molnar529565d2006-04-10 15:18:58 +02001403 if (off_out)
1404 return -ESPIPE;
Jens Axboeb92ce552006-04-11 13:52:07 +02001405 if (off_in) {
Changli Gao19c9a492010-06-29 13:10:36 +02001406 if (!(in->f_mode & FMODE_PREAD))
Jens Axboeb92ce552006-04-11 13:52:07 +02001407 return -EINVAL;
Jens Axboecbb7e572006-04-11 14:57:50 +02001408 if (copy_from_user(&offset, off_in, sizeof(loff_t)))
Jens Axboeb92ce552006-04-11 13:52:07 +02001409 return -EFAULT;
Al Viro7995bd22013-06-20 18:58:36 +04001410 } else {
1411 offset = in->f_pos;
1412 }
Ingo Molnar529565d2006-04-10 15:18:58 +02001413
Al Viro7995bd22013-06-20 18:58:36 +04001414 ret = do_splice_to(in, &offset, opipe, len, flags);
Jens Axboea4514eb2006-04-19 15:57:05 +02001415
Al Viro7995bd22013-06-20 18:58:36 +04001416 if (!off_in)
1417 in->f_pos = offset;
1418 else if (copy_to_user(off_in, &offset, sizeof(loff_t)))
Jens Axboea4514eb2006-04-19 15:57:05 +02001419 ret = -EFAULT;
1420
1421 return ret;
Ingo Molnar529565d2006-04-10 15:18:58 +02001422 }
Jens Axboe5274f052006-03-30 15:15:30 +02001423
1424 return -EINVAL;
1425}
1426
Jens Axboe912d35f2006-04-26 10:59:21 +02001427/*
1428 * Map an iov into an array of pages and offset/length tupples. With the
1429 * partial_page structure, we can map several non-contiguous ranges into
1430 * our ones pages[] map instead of splitting that operation into pieces.
1431 * Could easily be exported as a generic helper for other users, in which
1432 * case one would probably want to add a 'max_nr_pages' parameter as well.
1433 */
1434static int get_iovec_page_array(const struct iovec __user *iov,
1435 unsigned int nr_vecs, struct page **pages,
Eric Dumazetbd1a68b2012-04-04 11:08:51 +02001436 struct partial_page *partial, bool aligned,
Jens Axboe35f3d142010-05-20 10:43:18 +02001437 unsigned int pipe_buffers)
Jens Axboe912d35f2006-04-26 10:59:21 +02001438{
1439 int buffers = 0, error = 0;
1440
Jens Axboe912d35f2006-04-26 10:59:21 +02001441 while (nr_vecs) {
1442 unsigned long off, npages;
Linus Torvalds75723952007-10-01 13:17:28 -07001443 struct iovec entry;
Jens Axboe912d35f2006-04-26 10:59:21 +02001444 void __user *base;
1445 size_t len;
1446 int i;
1447
Linus Torvalds75723952007-10-01 13:17:28 -07001448 error = -EFAULT;
Nick Pigginbc40d732008-07-25 19:45:26 -07001449 if (copy_from_user(&entry, iov, sizeof(entry)))
Jens Axboe912d35f2006-04-26 10:59:21 +02001450 break;
Linus Torvalds75723952007-10-01 13:17:28 -07001451
1452 base = entry.iov_base;
1453 len = entry.iov_len;
Jens Axboe912d35f2006-04-26 10:59:21 +02001454
1455 /*
1456 * Sanity check this iovec. 0 read succeeds.
1457 */
Linus Torvalds75723952007-10-01 13:17:28 -07001458 error = 0;
Jens Axboe912d35f2006-04-26 10:59:21 +02001459 if (unlikely(!len))
1460 break;
1461 error = -EFAULT;
Bastian Blank712a30e2008-02-10 16:47:57 +02001462 if (!access_ok(VERIFY_READ, base, len))
Jens Axboe912d35f2006-04-26 10:59:21 +02001463 break;
1464
1465 /*
1466 * Get this base offset and number of pages, then map
1467 * in the user pages.
1468 */
1469 off = (unsigned long) base & ~PAGE_MASK;
Jens Axboe7afa6fd2006-05-01 20:02:33 +02001470
1471 /*
1472 * If asked for alignment, the offset must be zero and the
1473 * length a multiple of the PAGE_SIZE.
1474 */
1475 error = -EINVAL;
1476 if (aligned && (off || len & ~PAGE_MASK))
1477 break;
1478
Jens Axboe912d35f2006-04-26 10:59:21 +02001479 npages = (off + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
Jens Axboe35f3d142010-05-20 10:43:18 +02001480 if (npages > pipe_buffers - buffers)
1481 npages = pipe_buffers - buffers;
Jens Axboe912d35f2006-04-26 10:59:21 +02001482
Nick Pigginbc40d732008-07-25 19:45:26 -07001483 error = get_user_pages_fast((unsigned long)base, npages,
1484 0, &pages[buffers]);
Jens Axboe912d35f2006-04-26 10:59:21 +02001485
1486 if (unlikely(error <= 0))
1487 break;
1488
1489 /*
1490 * Fill this contiguous range into the partial page map.
1491 */
1492 for (i = 0; i < error; i++) {
Jens Axboe75914892006-05-02 12:57:18 +02001493 const int plen = min_t(size_t, len, PAGE_SIZE - off);
Jens Axboe912d35f2006-04-26 10:59:21 +02001494
1495 partial[buffers].offset = off;
1496 partial[buffers].len = plen;
1497
1498 off = 0;
1499 len -= plen;
1500 buffers++;
1501 }
1502
1503 /*
1504 * We didn't complete this iov, stop here since it probably
1505 * means we have to move some of this into a pipe to
1506 * be able to continue.
1507 */
1508 if (len)
1509 break;
1510
1511 /*
1512 * Don't continue if we mapped fewer pages than we asked for,
1513 * or if we mapped the max number of pages that we have
1514 * room for.
1515 */
Jens Axboe35f3d142010-05-20 10:43:18 +02001516 if (error < npages || buffers == pipe_buffers)
Jens Axboe912d35f2006-04-26 10:59:21 +02001517 break;
1518
1519 nr_vecs--;
1520 iov++;
1521 }
1522
Jens Axboe912d35f2006-04-26 10:59:21 +02001523 if (buffers)
1524 return buffers;
1525
1526 return error;
1527}
1528
Jens Axboe6a14b902007-06-14 13:08:55 +02001529static int pipe_to_user(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
1530 struct splice_desc *sd)
1531{
Al Viro6130f532014-02-03 18:19:51 -05001532 int n = copy_page_to_iter(buf->page, buf->offset, sd->len, sd->u.data);
1533 return n == sd->len ? n : -EFAULT;
Jens Axboe6a14b902007-06-14 13:08:55 +02001534}
1535
1536/*
1537 * For lack of a better implementation, implement vmsplice() to userspace
1538 * as a simple copy of the pipes pages to the user iov.
1539 */
Al Viro6130f532014-02-03 18:19:51 -05001540static long vmsplice_to_user(struct file *file, const struct iovec __user *uiov,
Jens Axboe6a14b902007-06-14 13:08:55 +02001541 unsigned long nr_segs, unsigned int flags)
1542{
1543 struct pipe_inode_info *pipe;
1544 struct splice_desc sd;
Jens Axboe6a14b902007-06-14 13:08:55 +02001545 long ret;
Al Viro6130f532014-02-03 18:19:51 -05001546 struct iovec iovstack[UIO_FASTIOV];
1547 struct iovec *iov = iovstack;
1548 struct iov_iter iter;
Jens Axboe6a14b902007-06-14 13:08:55 +02001549
Linus Torvalds71993e62010-11-28 13:56:09 -08001550 pipe = get_pipe_info(file);
Jens Axboe6a14b902007-06-14 13:08:55 +02001551 if (!pipe)
1552 return -EBADF;
1553
Al Viro345995f2015-03-21 19:17:55 -04001554 ret = import_iovec(READ, uiov, nr_segs,
1555 ARRAY_SIZE(iovstack), &iov, &iter);
1556 if (ret < 0)
1557 return ret;
Al Viro6130f532014-02-03 18:19:51 -05001558
Al Viro345995f2015-03-21 19:17:55 -04001559 sd.total_len = iov_iter_count(&iter);
Al Viro6130f532014-02-03 18:19:51 -05001560 sd.len = 0;
Al Viro6130f532014-02-03 18:19:51 -05001561 sd.flags = flags;
1562 sd.u.data = &iter;
1563 sd.pos = 0;
1564
Al Viro345995f2015-03-21 19:17:55 -04001565 if (sd.total_len) {
1566 pipe_lock(pipe);
1567 ret = __splice_from_pipe(pipe, &sd, pipe_to_user);
1568 pipe_unlock(pipe);
1569 }
Jens Axboe6a14b902007-06-14 13:08:55 +02001570
Al Viro345995f2015-03-21 19:17:55 -04001571 kfree(iov);
Jens Axboe6a14b902007-06-14 13:08:55 +02001572 return ret;
1573}
1574
Jens Axboe912d35f2006-04-26 10:59:21 +02001575/*
1576 * vmsplice splices a user address range into a pipe. It can be thought of
1577 * as splice-from-memory, where the regular splice is splice-from-file (or
1578 * to file). In both cases the output is a pipe, naturally.
Jens Axboe912d35f2006-04-26 10:59:21 +02001579 */
Jens Axboe6a14b902007-06-14 13:08:55 +02001580static long vmsplice_to_pipe(struct file *file, const struct iovec __user *iov,
1581 unsigned long nr_segs, unsigned int flags)
Jens Axboe912d35f2006-04-26 10:59:21 +02001582{
Jens Axboeddac0d32006-11-04 12:49:32 +01001583 struct pipe_inode_info *pipe;
Jens Axboe35f3d142010-05-20 10:43:18 +02001584 struct page *pages[PIPE_DEF_BUFFERS];
1585 struct partial_page partial[PIPE_DEF_BUFFERS];
Jens Axboe912d35f2006-04-26 10:59:21 +02001586 struct splice_pipe_desc spd = {
1587 .pages = pages,
1588 .partial = partial,
Eric Dumazet047fe362012-06-12 15:24:40 +02001589 .nr_pages_max = PIPE_DEF_BUFFERS,
Jens Axboe912d35f2006-04-26 10:59:21 +02001590 .flags = flags,
1591 .ops = &user_page_pipe_buf_ops,
Jens Axboebbdfc2f2007-11-06 23:29:47 -08001592 .spd_release = spd_release_page,
Jens Axboe912d35f2006-04-26 10:59:21 +02001593 };
Jens Axboe35f3d142010-05-20 10:43:18 +02001594 long ret;
Jens Axboe912d35f2006-04-26 10:59:21 +02001595
Linus Torvalds71993e62010-11-28 13:56:09 -08001596 pipe = get_pipe_info(file);
Jens Axboeddac0d32006-11-04 12:49:32 +01001597 if (!pipe)
Jens Axboe912d35f2006-04-26 10:59:21 +02001598 return -EBADF;
Jens Axboe912d35f2006-04-26 10:59:21 +02001599
Jens Axboe35f3d142010-05-20 10:43:18 +02001600 if (splice_grow_spd(pipe, &spd))
1601 return -ENOMEM;
Jens Axboe912d35f2006-04-26 10:59:21 +02001602
Jens Axboe35f3d142010-05-20 10:43:18 +02001603 spd.nr_pages = get_iovec_page_array(iov, nr_segs, spd.pages,
Eric Dumazetbd1a68b2012-04-04 11:08:51 +02001604 spd.partial, false,
Eric Dumazet047fe362012-06-12 15:24:40 +02001605 spd.nr_pages_max);
Jens Axboe35f3d142010-05-20 10:43:18 +02001606 if (spd.nr_pages <= 0)
1607 ret = spd.nr_pages;
1608 else
1609 ret = splice_to_pipe(pipe, &spd);
1610
Eric Dumazet047fe362012-06-12 15:24:40 +02001611 splice_shrink_spd(&spd);
Jens Axboe35f3d142010-05-20 10:43:18 +02001612 return ret;
Jens Axboe912d35f2006-04-26 10:59:21 +02001613}
1614
Jens Axboe6a14b902007-06-14 13:08:55 +02001615/*
1616 * Note that vmsplice only really supports true splicing _from_ user memory
1617 * to a pipe, not the other way around. Splicing from user memory is a simple
1618 * operation that can be supported without any funky alignment restrictions
1619 * or nasty vm tricks. We simply map in the user memory and fill them into
1620 * a pipe. The reverse isn't quite as easy, though. There are two possible
1621 * solutions for that:
1622 *
1623 * - memcpy() the data internally, at which point we might as well just
1624 * do a regular read() on the buffer anyway.
1625 * - Lots of nasty vm tricks, that are neither fast nor flexible (it
1626 * has restriction limitations on both ends of the pipe).
1627 *
1628 * Currently we punt and implement it as a normal copy, see pipe_to_user().
1629 *
1630 */
Heiko Carstens836f92a2009-01-14 14:14:33 +01001631SYSCALL_DEFINE4(vmsplice, int, fd, const struct iovec __user *, iov,
1632 unsigned long, nr_segs, unsigned int, flags)
Jens Axboe912d35f2006-04-26 10:59:21 +02001633{
Al Viro2903ff02012-08-28 12:52:22 -04001634 struct fd f;
Jens Axboe912d35f2006-04-26 10:59:21 +02001635 long error;
Jens Axboe912d35f2006-04-26 10:59:21 +02001636
Jens Axboe6a14b902007-06-14 13:08:55 +02001637 if (unlikely(nr_segs > UIO_MAXIOV))
1638 return -EINVAL;
1639 else if (unlikely(!nr_segs))
1640 return 0;
1641
Jens Axboe912d35f2006-04-26 10:59:21 +02001642 error = -EBADF;
Al Viro2903ff02012-08-28 12:52:22 -04001643 f = fdget(fd);
1644 if (f.file) {
1645 if (f.file->f_mode & FMODE_WRITE)
1646 error = vmsplice_to_pipe(f.file, iov, nr_segs, flags);
1647 else if (f.file->f_mode & FMODE_READ)
1648 error = vmsplice_to_user(f.file, iov, nr_segs, flags);
Jens Axboe912d35f2006-04-26 10:59:21 +02001649
Al Viro2903ff02012-08-28 12:52:22 -04001650 fdput(f);
Jens Axboe912d35f2006-04-26 10:59:21 +02001651 }
1652
1653 return error;
1654}
1655
Al Viro76b021d2013-03-02 10:19:56 -05001656#ifdef CONFIG_COMPAT
1657COMPAT_SYSCALL_DEFINE4(vmsplice, int, fd, const struct compat_iovec __user *, iov32,
1658 unsigned int, nr_segs, unsigned int, flags)
1659{
1660 unsigned i;
1661 struct iovec __user *iov;
1662 if (nr_segs > UIO_MAXIOV)
1663 return -EINVAL;
1664 iov = compat_alloc_user_space(nr_segs * sizeof(struct iovec));
1665 for (i = 0; i < nr_segs; i++) {
1666 struct compat_iovec v;
1667 if (get_user(v.iov_base, &iov32[i].iov_base) ||
1668 get_user(v.iov_len, &iov32[i].iov_len) ||
1669 put_user(compat_ptr(v.iov_base), &iov[i].iov_base) ||
1670 put_user(v.iov_len, &iov[i].iov_len))
1671 return -EFAULT;
1672 }
1673 return sys_vmsplice(fd, iov, nr_segs, flags);
1674}
1675#endif
1676
Heiko Carstens836f92a2009-01-14 14:14:33 +01001677SYSCALL_DEFINE6(splice, int, fd_in, loff_t __user *, off_in,
1678 int, fd_out, loff_t __user *, off_out,
1679 size_t, len, unsigned int, flags)
Jens Axboe5274f052006-03-30 15:15:30 +02001680{
Al Viro2903ff02012-08-28 12:52:22 -04001681 struct fd in, out;
Jens Axboe5274f052006-03-30 15:15:30 +02001682 long error;
Jens Axboe5274f052006-03-30 15:15:30 +02001683
1684 if (unlikely(!len))
1685 return 0;
1686
1687 error = -EBADF;
Al Viro2903ff02012-08-28 12:52:22 -04001688 in = fdget(fd_in);
1689 if (in.file) {
1690 if (in.file->f_mode & FMODE_READ) {
1691 out = fdget(fd_out);
1692 if (out.file) {
1693 if (out.file->f_mode & FMODE_WRITE)
1694 error = do_splice(in.file, off_in,
1695 out.file, off_out,
Ingo Molnar529565d2006-04-10 15:18:58 +02001696 len, flags);
Al Viro2903ff02012-08-28 12:52:22 -04001697 fdput(out);
Jens Axboe5274f052006-03-30 15:15:30 +02001698 }
1699 }
Al Viro2903ff02012-08-28 12:52:22 -04001700 fdput(in);
Jens Axboe5274f052006-03-30 15:15:30 +02001701 }
Jens Axboe5274f052006-03-30 15:15:30 +02001702 return error;
1703}
Jens Axboe70524492006-04-11 15:51:17 +02001704
1705/*
Jens Axboeaadd06e2006-07-10 11:00:01 +02001706 * Make sure there's data to read. Wait for input if we can, otherwise
1707 * return an appropriate error.
1708 */
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001709static int ipipe_prep(struct pipe_inode_info *pipe, unsigned int flags)
Jens Axboeaadd06e2006-07-10 11:00:01 +02001710{
1711 int ret;
1712
1713 /*
1714 * Check ->nrbufs without the inode lock first. This function
1715 * is speculative anyways, so missing one is ok.
1716 */
1717 if (pipe->nrbufs)
1718 return 0;
1719
1720 ret = 0;
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001721 pipe_lock(pipe);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001722
1723 while (!pipe->nrbufs) {
1724 if (signal_pending(current)) {
1725 ret = -ERESTARTSYS;
1726 break;
1727 }
1728 if (!pipe->writers)
1729 break;
1730 if (!pipe->waiting_writers) {
1731 if (flags & SPLICE_F_NONBLOCK) {
1732 ret = -EAGAIN;
1733 break;
1734 }
1735 }
1736 pipe_wait(pipe);
1737 }
1738
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001739 pipe_unlock(pipe);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001740 return ret;
1741}
1742
1743/*
1744 * Make sure there's writeable room. Wait for room if we can, otherwise
1745 * return an appropriate error.
1746 */
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001747static int opipe_prep(struct pipe_inode_info *pipe, unsigned int flags)
Jens Axboeaadd06e2006-07-10 11:00:01 +02001748{
1749 int ret;
1750
1751 /*
1752 * Check ->nrbufs without the inode lock first. This function
1753 * is speculative anyways, so missing one is ok.
1754 */
Jens Axboe35f3d142010-05-20 10:43:18 +02001755 if (pipe->nrbufs < pipe->buffers)
Jens Axboeaadd06e2006-07-10 11:00:01 +02001756 return 0;
1757
1758 ret = 0;
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001759 pipe_lock(pipe);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001760
Jens Axboe35f3d142010-05-20 10:43:18 +02001761 while (pipe->nrbufs >= pipe->buffers) {
Jens Axboeaadd06e2006-07-10 11:00:01 +02001762 if (!pipe->readers) {
1763 send_sig(SIGPIPE, current, 0);
1764 ret = -EPIPE;
1765 break;
1766 }
1767 if (flags & SPLICE_F_NONBLOCK) {
1768 ret = -EAGAIN;
1769 break;
1770 }
1771 if (signal_pending(current)) {
1772 ret = -ERESTARTSYS;
1773 break;
1774 }
1775 pipe->waiting_writers++;
1776 pipe_wait(pipe);
1777 pipe->waiting_writers--;
1778 }
1779
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001780 pipe_unlock(pipe);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001781 return ret;
1782}
1783
1784/*
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001785 * Splice contents of ipipe to opipe.
1786 */
1787static int splice_pipe_to_pipe(struct pipe_inode_info *ipipe,
1788 struct pipe_inode_info *opipe,
1789 size_t len, unsigned int flags)
1790{
1791 struct pipe_buffer *ibuf, *obuf;
1792 int ret = 0, nbuf;
1793 bool input_wakeup = false;
1794
1795
1796retry:
1797 ret = ipipe_prep(ipipe, flags);
1798 if (ret)
1799 return ret;
1800
1801 ret = opipe_prep(opipe, flags);
1802 if (ret)
1803 return ret;
1804
1805 /*
1806 * Potential ABBA deadlock, work around it by ordering lock
1807 * grabbing by pipe info address. Otherwise two different processes
1808 * could deadlock (one doing tee from A -> B, the other from B -> A).
1809 */
1810 pipe_double_lock(ipipe, opipe);
1811
1812 do {
1813 if (!opipe->readers) {
1814 send_sig(SIGPIPE, current, 0);
1815 if (!ret)
1816 ret = -EPIPE;
1817 break;
1818 }
1819
1820 if (!ipipe->nrbufs && !ipipe->writers)
1821 break;
1822
1823 /*
1824 * Cannot make any progress, because either the input
1825 * pipe is empty or the output pipe is full.
1826 */
Jens Axboe35f3d142010-05-20 10:43:18 +02001827 if (!ipipe->nrbufs || opipe->nrbufs >= opipe->buffers) {
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001828 /* Already processed some buffers, break */
1829 if (ret)
1830 break;
1831
1832 if (flags & SPLICE_F_NONBLOCK) {
1833 ret = -EAGAIN;
1834 break;
1835 }
1836
1837 /*
1838 * We raced with another reader/writer and haven't
1839 * managed to process any buffers. A zero return
1840 * value means EOF, so retry instead.
1841 */
1842 pipe_unlock(ipipe);
1843 pipe_unlock(opipe);
1844 goto retry;
1845 }
1846
1847 ibuf = ipipe->bufs + ipipe->curbuf;
Jens Axboe35f3d142010-05-20 10:43:18 +02001848 nbuf = (opipe->curbuf + opipe->nrbufs) & (opipe->buffers - 1);
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001849 obuf = opipe->bufs + nbuf;
1850
1851 if (len >= ibuf->len) {
1852 /*
1853 * Simply move the whole buffer from ipipe to opipe
1854 */
1855 *obuf = *ibuf;
1856 ibuf->ops = NULL;
1857 opipe->nrbufs++;
Jens Axboe35f3d142010-05-20 10:43:18 +02001858 ipipe->curbuf = (ipipe->curbuf + 1) & (ipipe->buffers - 1);
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001859 ipipe->nrbufs--;
1860 input_wakeup = true;
1861 } else {
1862 /*
1863 * Get a reference to this pipe buffer,
1864 * so we can copy the contents over.
1865 */
1866 ibuf->ops->get(ipipe, ibuf);
1867 *obuf = *ibuf;
1868
1869 /*
1870 * Don't inherit the gift flag, we need to
1871 * prevent multiple steals of this page.
1872 */
1873 obuf->flags &= ~PIPE_BUF_FLAG_GIFT;
1874
1875 obuf->len = len;
1876 opipe->nrbufs++;
1877 ibuf->offset += obuf->len;
1878 ibuf->len -= obuf->len;
1879 }
1880 ret += obuf->len;
1881 len -= obuf->len;
1882 } while (len);
1883
1884 pipe_unlock(ipipe);
1885 pipe_unlock(opipe);
1886
1887 /*
1888 * If we put data in the output pipe, wakeup any potential readers.
1889 */
Namhyung Kim825cdcb2011-05-23 19:58:53 +02001890 if (ret > 0)
1891 wakeup_pipe_readers(opipe);
1892
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001893 if (input_wakeup)
1894 wakeup_pipe_writers(ipipe);
1895
1896 return ret;
1897}
1898
1899/*
Jens Axboe70524492006-04-11 15:51:17 +02001900 * Link contents of ipipe to opipe.
1901 */
1902static int link_pipe(struct pipe_inode_info *ipipe,
1903 struct pipe_inode_info *opipe,
1904 size_t len, unsigned int flags)
1905{
1906 struct pipe_buffer *ibuf, *obuf;
Jens Axboeaadd06e2006-07-10 11:00:01 +02001907 int ret = 0, i = 0, nbuf;
Jens Axboe70524492006-04-11 15:51:17 +02001908
1909 /*
1910 * Potential ABBA deadlock, work around it by ordering lock
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001911 * grabbing by pipe info address. Otherwise two different processes
Jens Axboe70524492006-04-11 15:51:17 +02001912 * could deadlock (one doing tee from A -> B, the other from B -> A).
1913 */
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001914 pipe_double_lock(ipipe, opipe);
Jens Axboe70524492006-04-11 15:51:17 +02001915
Jens Axboeaadd06e2006-07-10 11:00:01 +02001916 do {
Jens Axboe70524492006-04-11 15:51:17 +02001917 if (!opipe->readers) {
1918 send_sig(SIGPIPE, current, 0);
1919 if (!ret)
1920 ret = -EPIPE;
1921 break;
1922 }
Jens Axboe70524492006-04-11 15:51:17 +02001923
1924 /*
Jens Axboeaadd06e2006-07-10 11:00:01 +02001925 * If we have iterated all input buffers or ran out of
1926 * output room, break.
Jens Axboe70524492006-04-11 15:51:17 +02001927 */
Jens Axboe35f3d142010-05-20 10:43:18 +02001928 if (i >= ipipe->nrbufs || opipe->nrbufs >= opipe->buffers)
Jens Axboe70524492006-04-11 15:51:17 +02001929 break;
Jens Axboeaadd06e2006-07-10 11:00:01 +02001930
Jens Axboe35f3d142010-05-20 10:43:18 +02001931 ibuf = ipipe->bufs + ((ipipe->curbuf + i) & (ipipe->buffers-1));
1932 nbuf = (opipe->curbuf + opipe->nrbufs) & (opipe->buffers - 1);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001933
Jens Axboe2a27250e2006-04-19 15:56:40 +02001934 /*
Jens Axboeaadd06e2006-07-10 11:00:01 +02001935 * Get a reference to this pipe buffer,
1936 * so we can copy the contents over.
Jens Axboe2a27250e2006-04-19 15:56:40 +02001937 */
Jens Axboeaadd06e2006-07-10 11:00:01 +02001938 ibuf->ops->get(ipipe, ibuf);
Jens Axboe70524492006-04-11 15:51:17 +02001939
Jens Axboeaadd06e2006-07-10 11:00:01 +02001940 obuf = opipe->bufs + nbuf;
1941 *obuf = *ibuf;
Jens Axboe70524492006-04-11 15:51:17 +02001942
Jens Axboeaadd06e2006-07-10 11:00:01 +02001943 /*
1944 * Don't inherit the gift flag, we need to
1945 * prevent multiple steals of this page.
1946 */
1947 obuf->flags &= ~PIPE_BUF_FLAG_GIFT;
1948
1949 if (obuf->len > len)
1950 obuf->len = len;
1951
1952 opipe->nrbufs++;
1953 ret += obuf->len;
1954 len -= obuf->len;
1955 i++;
1956 } while (len);
Jens Axboe70524492006-04-11 15:51:17 +02001957
Jens Axboe02cf01a2008-02-20 10:34:51 +01001958 /*
1959 * return EAGAIN if we have the potential of some data in the
1960 * future, otherwise just return 0
1961 */
1962 if (!ret && ipipe->waiting_writers && (flags & SPLICE_F_NONBLOCK))
1963 ret = -EAGAIN;
1964
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001965 pipe_unlock(ipipe);
1966 pipe_unlock(opipe);
Jens Axboe70524492006-04-11 15:51:17 +02001967
Jens Axboeaadd06e2006-07-10 11:00:01 +02001968 /*
1969 * If we put data in the output pipe, wakeup any potential readers.
1970 */
Namhyung Kim825cdcb2011-05-23 19:58:53 +02001971 if (ret > 0)
1972 wakeup_pipe_readers(opipe);
Jens Axboe70524492006-04-11 15:51:17 +02001973
1974 return ret;
1975}
1976
1977/*
1978 * This is a tee(1) implementation that works on pipes. It doesn't copy
1979 * any data, it simply references the 'in' pages on the 'out' pipe.
1980 * The 'flags' used are the SPLICE_F_* variants, currently the only
1981 * applicable one is SPLICE_F_NONBLOCK.
1982 */
1983static long do_tee(struct file *in, struct file *out, size_t len,
1984 unsigned int flags)
1985{
Linus Torvalds71993e62010-11-28 13:56:09 -08001986 struct pipe_inode_info *ipipe = get_pipe_info(in);
1987 struct pipe_inode_info *opipe = get_pipe_info(out);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001988 int ret = -EINVAL;
Jens Axboe70524492006-04-11 15:51:17 +02001989
1990 /*
Jens Axboeaadd06e2006-07-10 11:00:01 +02001991 * Duplicate the contents of ipipe to opipe without actually
1992 * copying the data.
Jens Axboe70524492006-04-11 15:51:17 +02001993 */
Jens Axboeaadd06e2006-07-10 11:00:01 +02001994 if (ipipe && opipe && ipipe != opipe) {
1995 /*
1996 * Keep going, unless we encounter an error. The ipipe/opipe
1997 * ordering doesn't really matter.
1998 */
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001999 ret = ipipe_prep(ipipe, flags);
Jens Axboeaadd06e2006-07-10 11:00:01 +02002000 if (!ret) {
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02002001 ret = opipe_prep(opipe, flags);
Jens Axboe02cf01a2008-02-20 10:34:51 +01002002 if (!ret)
Jens Axboeaadd06e2006-07-10 11:00:01 +02002003 ret = link_pipe(ipipe, opipe, len, flags);
Jens Axboeaadd06e2006-07-10 11:00:01 +02002004 }
2005 }
Jens Axboe70524492006-04-11 15:51:17 +02002006
Jens Axboeaadd06e2006-07-10 11:00:01 +02002007 return ret;
Jens Axboe70524492006-04-11 15:51:17 +02002008}
2009
Heiko Carstens836f92a2009-01-14 14:14:33 +01002010SYSCALL_DEFINE4(tee, int, fdin, int, fdout, size_t, len, unsigned int, flags)
Jens Axboe70524492006-04-11 15:51:17 +02002011{
Al Viro2903ff02012-08-28 12:52:22 -04002012 struct fd in;
2013 int error;
Jens Axboe70524492006-04-11 15:51:17 +02002014
2015 if (unlikely(!len))
2016 return 0;
2017
2018 error = -EBADF;
Al Viro2903ff02012-08-28 12:52:22 -04002019 in = fdget(fdin);
2020 if (in.file) {
2021 if (in.file->f_mode & FMODE_READ) {
2022 struct fd out = fdget(fdout);
2023 if (out.file) {
2024 if (out.file->f_mode & FMODE_WRITE)
2025 error = do_tee(in.file, out.file,
2026 len, flags);
2027 fdput(out);
Jens Axboe70524492006-04-11 15:51:17 +02002028 }
2029 }
Al Viro2903ff02012-08-28 12:52:22 -04002030 fdput(in);
Jens Axboe70524492006-04-11 15:51:17 +02002031 }
2032
2033 return error;
2034}