blob: e246954ea48cb486b1c8101e6f621364d6844535 [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 Wu431a48202007-07-19 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 Wu431a48202007-07-19 01:48:05 -0700340 * readahead/allocate the rest and fill in the holes.
Jens Axboeeb207962006-04-27 11:05:22 +0200341 */
Fengguang Wu431a48202007-07-19 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 Wu431a48202007-07-19 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
526 isize = i_size_read(in->f_mapping->host);
527 if (unlikely(*ppos >= isize))
528 return 0;
529
530 left = isize - *ppos;
531 if (unlikely(left < len))
532 len = left;
Jens Axboe5274f052006-03-30 15:15:30 +0200533
Jens Axboe8191ecd2008-04-10 08:24:25 +0200534 ret = __generic_file_splice_read(in, ppos, pipe, len, flags);
Miklos Szeredi723590e2009-08-15 08:43:22 +0200535 if (ret > 0) {
Jens Axboecbb7e572006-04-11 14:57:50 +0200536 *ppos += ret;
Miklos Szeredi723590e2009-08-15 08:43:22 +0200537 file_accessed(in);
538 }
Jens Axboe5274f052006-03-30 15:15:30 +0200539
540 return ret;
541}
Jens Axboe059a8f32006-04-02 23:06:05 +0200542EXPORT_SYMBOL(generic_file_splice_read);
543
Miklos Szeredi68181732009-05-07 15:37:36 +0200544static const struct pipe_buf_operations default_pipe_buf_ops = {
545 .can_merge = 0,
Miklos Szeredi68181732009-05-07 15:37:36 +0200546 .confirm = generic_pipe_buf_confirm,
547 .release = generic_pipe_buf_release,
548 .steal = generic_pipe_buf_steal,
549 .get = generic_pipe_buf_get,
550};
551
Miklos Szeredi28a625c2014-01-22 19:36:57 +0100552static int generic_pipe_buf_nosteal(struct pipe_inode_info *pipe,
553 struct pipe_buffer *buf)
554{
555 return 1;
556}
557
558/* Pipe buffer operations for a socket and similar. */
559const struct pipe_buf_operations nosteal_pipe_buf_ops = {
560 .can_merge = 0,
Miklos Szeredi28a625c2014-01-22 19:36:57 +0100561 .confirm = generic_pipe_buf_confirm,
562 .release = generic_pipe_buf_release,
563 .steal = generic_pipe_buf_nosteal,
564 .get = generic_pipe_buf_get,
565};
566EXPORT_SYMBOL(nosteal_pipe_buf_ops);
567
Miklos Szeredi68181732009-05-07 15:37:36 +0200568static ssize_t kernel_readv(struct file *file, const struct iovec *vec,
569 unsigned long vlen, loff_t offset)
570{
571 mm_segment_t old_fs;
572 loff_t pos = offset;
573 ssize_t res;
574
575 old_fs = get_fs();
576 set_fs(get_ds());
577 /* The cast to a user pointer is valid due to the set_fs() */
578 res = vfs_readv(file, (const struct iovec __user *)vec, vlen, &pos);
579 set_fs(old_fs);
580
581 return res;
582}
583
Al Viro7bb307e2013-02-23 14:51:48 -0500584ssize_t kernel_write(struct file *file, const char *buf, size_t count,
Miklos Szeredib2858d72009-05-19 11:37:46 +0200585 loff_t pos)
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +0200586{
587 mm_segment_t old_fs;
588 ssize_t res;
589
590 old_fs = get_fs();
591 set_fs(get_ds());
592 /* The cast to a user pointer is valid due to the set_fs() */
Al Viro7bb307e2013-02-23 14:51:48 -0500593 res = vfs_write(file, (__force const char __user *)buf, count, &pos);
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +0200594 set_fs(old_fs);
595
596 return res;
597}
Al Viro7bb307e2013-02-23 14:51:48 -0500598EXPORT_SYMBOL(kernel_write);
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +0200599
Miklos Szeredi68181732009-05-07 15:37:36 +0200600ssize_t default_file_splice_read(struct file *in, loff_t *ppos,
601 struct pipe_inode_info *pipe, size_t len,
602 unsigned int flags)
603{
604 unsigned int nr_pages;
605 unsigned int nr_freed;
606 size_t offset;
Jens Axboe35f3d142010-05-20 10:43:18 +0200607 struct page *pages[PIPE_DEF_BUFFERS];
608 struct partial_page partial[PIPE_DEF_BUFFERS];
609 struct iovec *vec, __vec[PIPE_DEF_BUFFERS];
Miklos Szeredi68181732009-05-07 15:37:36 +0200610 ssize_t res;
611 size_t this_len;
612 int error;
613 int i;
614 struct splice_pipe_desc spd = {
615 .pages = pages,
616 .partial = partial,
Eric Dumazet047fe362012-06-12 15:24:40 +0200617 .nr_pages_max = PIPE_DEF_BUFFERS,
Miklos Szeredi68181732009-05-07 15:37:36 +0200618 .flags = flags,
619 .ops = &default_pipe_buf_ops,
620 .spd_release = spd_release_page,
621 };
622
Jens Axboe35f3d142010-05-20 10:43:18 +0200623 if (splice_grow_spd(pipe, &spd))
624 return -ENOMEM;
625
626 res = -ENOMEM;
627 vec = __vec;
Eric Dumazet047fe362012-06-12 15:24:40 +0200628 if (spd.nr_pages_max > PIPE_DEF_BUFFERS) {
629 vec = kmalloc(spd.nr_pages_max * sizeof(struct iovec), GFP_KERNEL);
Jens Axboe35f3d142010-05-20 10:43:18 +0200630 if (!vec)
631 goto shrink_ret;
632 }
633
Miklos Szeredi68181732009-05-07 15:37:36 +0200634 offset = *ppos & ~PAGE_CACHE_MASK;
635 nr_pages = (len + offset + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
636
Eric Dumazet047fe362012-06-12 15:24:40 +0200637 for (i = 0; i < nr_pages && i < spd.nr_pages_max && len; i++) {
Miklos Szeredi68181732009-05-07 15:37:36 +0200638 struct page *page;
639
Jens Axboe4f231222009-05-13 08:35:35 +0200640 page = alloc_page(GFP_USER);
Miklos Szeredi68181732009-05-07 15:37:36 +0200641 error = -ENOMEM;
642 if (!page)
643 goto err;
644
645 this_len = min_t(size_t, len, PAGE_CACHE_SIZE - offset);
Jens Axboe4f231222009-05-13 08:35:35 +0200646 vec[i].iov_base = (void __user *) page_address(page);
Miklos Szeredi68181732009-05-07 15:37:36 +0200647 vec[i].iov_len = this_len;
Jens Axboe35f3d142010-05-20 10:43:18 +0200648 spd.pages[i] = page;
Miklos Szeredi68181732009-05-07 15:37:36 +0200649 spd.nr_pages++;
650 len -= this_len;
651 offset = 0;
652 }
653
654 res = kernel_readv(in, vec, spd.nr_pages, *ppos);
Andrew Morton77f6bf52009-05-14 09:49:44 +0200655 if (res < 0) {
656 error = res;
Miklos Szeredi68181732009-05-07 15:37:36 +0200657 goto err;
Andrew Morton77f6bf52009-05-14 09:49:44 +0200658 }
Miklos Szeredi68181732009-05-07 15:37:36 +0200659
660 error = 0;
661 if (!res)
662 goto err;
663
664 nr_freed = 0;
665 for (i = 0; i < spd.nr_pages; i++) {
Miklos Szeredi68181732009-05-07 15:37:36 +0200666 this_len = min_t(size_t, vec[i].iov_len, res);
Jens Axboe35f3d142010-05-20 10:43:18 +0200667 spd.partial[i].offset = 0;
668 spd.partial[i].len = this_len;
Miklos Szeredi68181732009-05-07 15:37:36 +0200669 if (!this_len) {
Jens Axboe35f3d142010-05-20 10:43:18 +0200670 __free_page(spd.pages[i]);
671 spd.pages[i] = NULL;
Miklos Szeredi68181732009-05-07 15:37:36 +0200672 nr_freed++;
673 }
674 res -= this_len;
675 }
676 spd.nr_pages -= nr_freed;
677
678 res = splice_to_pipe(pipe, &spd);
679 if (res > 0)
680 *ppos += res;
681
Jens Axboe35f3d142010-05-20 10:43:18 +0200682shrink_ret:
683 if (vec != __vec)
684 kfree(vec);
Eric Dumazet047fe362012-06-12 15:24:40 +0200685 splice_shrink_spd(&spd);
Miklos Szeredi68181732009-05-07 15:37:36 +0200686 return res;
687
688err:
Jens Axboe4f231222009-05-13 08:35:35 +0200689 for (i = 0; i < spd.nr_pages; i++)
Jens Axboe35f3d142010-05-20 10:43:18 +0200690 __free_page(spd.pages[i]);
Jens Axboe4f231222009-05-13 08:35:35 +0200691
Jens Axboe35f3d142010-05-20 10:43:18 +0200692 res = error;
693 goto shrink_ret;
Miklos Szeredi68181732009-05-07 15:37:36 +0200694}
695EXPORT_SYMBOL(default_file_splice_read);
696
Jens Axboe5274f052006-03-30 15:15:30 +0200697/*
Jens Axboe4f6f0bd2006-04-02 23:04:46 +0200698 * Send 'sd->len' bytes to socket from 'sd->file' at position 'sd->pos'
Jens Axboe016b6612006-04-25 15:42:00 +0200699 * using sendpage(). Return the number of bytes sent.
Jens Axboe5274f052006-03-30 15:15:30 +0200700 */
Jens Axboe76ad4d12006-05-03 10:41:33 +0200701static int pipe_to_sendpage(struct pipe_inode_info *pipe,
Jens Axboe5274f052006-03-30 15:15:30 +0200702 struct pipe_buffer *buf, struct splice_desc *sd)
703{
Jens Axboe6a14b902007-06-14 13:08:55 +0200704 struct file *file = sd->u.file;
Jens Axboe5274f052006-03-30 15:15:30 +0200705 loff_t pos = sd->pos;
Michał Mirosława8adbe32010-12-17 08:56:44 +0100706 int more;
Jens Axboe5274f052006-03-30 15:15:30 +0200707
Al Viro72c2d532013-09-22 16:27:52 -0400708 if (!likely(file->f_op->sendpage))
Michał Mirosława8adbe32010-12-17 08:56:44 +0100709 return -EINVAL;
Jens Axboe5274f052006-03-30 15:15:30 +0200710
Eric Dumazet35f9c092012-04-05 03:05:35 +0000711 more = (sd->flags & SPLICE_F_MORE) ? MSG_MORE : 0;
Eric Dumazetae62ca72013-01-06 18:21:49 +0000712
713 if (sd->len < sd->total_len && pipe->nrbufs > 1)
Eric Dumazet35f9c092012-04-05 03:05:35 +0000714 more |= MSG_SENDPAGE_NOTLAST;
Eric Dumazetae62ca72013-01-06 18:21:49 +0000715
Michał Mirosława8adbe32010-12-17 08:56:44 +0100716 return file->f_op->sendpage(file, buf->page, buf->offset,
717 sd->len, &pos, more);
Jens Axboe5274f052006-03-30 15:15:30 +0200718}
719
720/*
721 * This is a little more tricky than the file -> pipe splicing. There are
722 * basically three cases:
723 *
724 * - Destination page already exists in the address space and there
725 * are users of it. For that case we have no other option that
726 * copying the data. Tough luck.
727 * - Destination page already exists in the address space, but there
728 * are no users of it. Make sure it's uptodate, then drop it. Fall
729 * through to last case.
730 * - Destination page does not exist, we can add the pipe page to
731 * the page cache and avoid the copy.
732 *
Jens Axboe83f91352006-04-02 23:05:09 +0200733 * If asked to move pages to the output file (SPLICE_F_MOVE is set in
734 * sd->flags), we attempt to migrate pages from the pipe to the output
735 * file address space page cache. This is possible if no one else has
736 * the pipe page referenced outside of the pipe and page cache. If
737 * SPLICE_F_MOVE isn't set, or we cannot move the page, we simply create
738 * a new page in the output file page cache and fill/dirty that.
Jens Axboe5274f052006-03-30 15:15:30 +0200739 */
Miklos Szeredi328eaab2009-04-14 19:48:39 +0200740int pipe_to_file(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
741 struct splice_desc *sd)
Jens Axboe5274f052006-03-30 15:15:30 +0200742{
Jens Axboe6a14b902007-06-14 13:08:55 +0200743 struct file *file = sd->u.file;
Jens Axboe5274f052006-03-30 15:15:30 +0200744 struct address_space *mapping = file->f_mapping;
Jens Axboe016b6612006-04-25 15:42:00 +0200745 unsigned int offset, this_len;
Jens Axboe5274f052006-03-30 15:15:30 +0200746 struct page *page;
Nick Pigginafddba42007-10-16 01:25:01 -0700747 void *fsdata;
Jens Axboe3e7ee3e2006-04-02 23:11:04 +0200748 int ret;
Jens Axboe5274f052006-03-30 15:15:30 +0200749
Jens Axboe5274f052006-03-30 15:15:30 +0200750 offset = sd->pos & ~PAGE_CACHE_MASK;
751
Jens Axboe016b6612006-04-25 15:42:00 +0200752 this_len = sd->len;
753 if (this_len + offset > PAGE_CACHE_SIZE)
754 this_len = PAGE_CACHE_SIZE - offset;
755
Nick Pigginafddba42007-10-16 01:25:01 -0700756 ret = pagecache_write_begin(file, mapping, sd->pos, this_len,
757 AOP_FLAG_UNINTERRUPTIBLE, &page, &fsdata);
758 if (unlikely(ret))
759 goto out;
Jens Axboe5274f052006-03-30 15:15:30 +0200760
Jens Axboe0568b402006-05-01 19:50:48 +0200761 if (buf->page != page) {
Al Virofbb32752014-02-02 21:09:54 -0500762 char *src = kmap_atomic(buf->page);
Cong Wange8e3c3d2011-11-25 23:14:27 +0800763 char *dst = kmap_atomic(page);
Jens Axboe5abc97a2006-03-30 15:16:46 +0200764
Jens Axboe016b6612006-04-25 15:42:00 +0200765 memcpy(dst + offset, src + buf->offset, this_len);
Jens Axboe5abc97a2006-03-30 15:16:46 +0200766 flush_dcache_page(page);
Cong Wange8e3c3d2011-11-25 23:14:27 +0800767 kunmap_atomic(dst);
Al Virofbb32752014-02-02 21:09:54 -0500768 kunmap_atomic(src);
Jens Axboe5abc97a2006-03-30 15:16:46 +0200769 }
Nick Pigginafddba42007-10-16 01:25:01 -0700770 ret = pagecache_write_end(file, mapping, sd->pos, this_len, this_len,
771 page, fsdata);
Jens Axboe5274f052006-03-30 15:15:30 +0200772out:
Jens Axboe5274f052006-03-30 15:15:30 +0200773 return ret;
774}
Miklos Szeredi328eaab2009-04-14 19:48:39 +0200775EXPORT_SYMBOL(pipe_to_file);
Jens Axboe5274f052006-03-30 15:15:30 +0200776
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200777static void wakeup_pipe_writers(struct pipe_inode_info *pipe)
778{
779 smp_mb();
780 if (waitqueue_active(&pipe->wait))
781 wake_up_interruptible(&pipe->wait);
782 kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
783}
784
785/**
786 * splice_from_pipe_feed - feed available data from a pipe to a file
787 * @pipe: pipe to splice from
788 * @sd: information to @actor
789 * @actor: handler that splices the data
790 *
791 * Description:
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200792 * This function loops over the pipe and calls @actor to do the
793 * actual moving of a single struct pipe_buffer to the desired
794 * destination. It returns when there's no more buffers left in
795 * the pipe or if the requested number of bytes (@sd->total_len)
796 * have been copied. It returns a positive number (one) if the
797 * pipe needs to be filled with more data, zero if the required
798 * number of bytes have been copied and -errno on error.
799 *
800 * This, together with splice_from_pipe_{begin,end,next}, may be
801 * used to implement the functionality of __splice_from_pipe() when
802 * locking is required around copying the pipe buffers to the
803 * destination.
804 */
805int splice_from_pipe_feed(struct pipe_inode_info *pipe, struct splice_desc *sd,
806 splice_actor *actor)
807{
808 int ret;
809
810 while (pipe->nrbufs) {
811 struct pipe_buffer *buf = pipe->bufs + pipe->curbuf;
812 const struct pipe_buf_operations *ops = buf->ops;
813
814 sd->len = buf->len;
815 if (sd->len > sd->total_len)
816 sd->len = sd->total_len;
817
Michał Mirosława8adbe32010-12-17 08:56:44 +0100818 ret = buf->ops->confirm(pipe, buf);
819 if (unlikely(ret)) {
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200820 if (ret == -ENODATA)
821 ret = 0;
822 return ret;
823 }
Michał Mirosława8adbe32010-12-17 08:56:44 +0100824
825 ret = actor(pipe, buf, sd);
826 if (ret <= 0)
827 return ret;
828
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200829 buf->offset += ret;
830 buf->len -= ret;
831
832 sd->num_spliced += ret;
833 sd->len -= ret;
834 sd->pos += ret;
835 sd->total_len -= ret;
836
837 if (!buf->len) {
838 buf->ops = NULL;
839 ops->release(pipe, buf);
Jens Axboe35f3d142010-05-20 10:43:18 +0200840 pipe->curbuf = (pipe->curbuf + 1) & (pipe->buffers - 1);
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200841 pipe->nrbufs--;
Al Viro6447a3c2013-03-21 11:01:38 -0400842 if (pipe->files)
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200843 sd->need_wakeup = true;
844 }
845
846 if (!sd->total_len)
847 return 0;
848 }
849
850 return 1;
851}
852EXPORT_SYMBOL(splice_from_pipe_feed);
853
854/**
855 * splice_from_pipe_next - wait for some data to splice from
856 * @pipe: pipe to splice from
857 * @sd: information about the splice operation
858 *
859 * Description:
860 * This function will wait for some data and return a positive
861 * value (one) if pipe buffers are available. It will return zero
862 * or -errno if no more data needs to be spliced.
863 */
864int splice_from_pipe_next(struct pipe_inode_info *pipe, struct splice_desc *sd)
865{
866 while (!pipe->nrbufs) {
867 if (!pipe->writers)
868 return 0;
869
870 if (!pipe->waiting_writers && sd->num_spliced)
871 return 0;
872
873 if (sd->flags & SPLICE_F_NONBLOCK)
874 return -EAGAIN;
875
876 if (signal_pending(current))
877 return -ERESTARTSYS;
878
879 if (sd->need_wakeup) {
880 wakeup_pipe_writers(pipe);
881 sd->need_wakeup = false;
882 }
883
884 pipe_wait(pipe);
885 }
886
887 return 1;
888}
889EXPORT_SYMBOL(splice_from_pipe_next);
890
891/**
892 * splice_from_pipe_begin - start splicing from pipe
Randy Dunlapb80901b2009-04-16 19:09:55 -0700893 * @sd: information about the splice operation
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200894 *
895 * Description:
896 * This function should be called before a loop containing
897 * splice_from_pipe_next() and splice_from_pipe_feed() to
898 * initialize the necessary fields of @sd.
899 */
900void splice_from_pipe_begin(struct splice_desc *sd)
901{
902 sd->num_spliced = 0;
903 sd->need_wakeup = false;
904}
905EXPORT_SYMBOL(splice_from_pipe_begin);
906
907/**
908 * splice_from_pipe_end - finish splicing from pipe
909 * @pipe: pipe to splice from
910 * @sd: information about the splice operation
911 *
912 * Description:
913 * This function will wake up pipe writers if necessary. It should
914 * be called after a loop containing splice_from_pipe_next() and
915 * splice_from_pipe_feed().
916 */
917void splice_from_pipe_end(struct pipe_inode_info *pipe, struct splice_desc *sd)
918{
919 if (sd->need_wakeup)
920 wakeup_pipe_writers(pipe);
921}
922EXPORT_SYMBOL(splice_from_pipe_end);
923
Jens Axboe932cc6d2007-06-21 13:10:21 +0200924/**
925 * __splice_from_pipe - splice data from a pipe to given actor
926 * @pipe: pipe to splice from
927 * @sd: information to @actor
928 * @actor: handler that splices the data
929 *
930 * Description:
931 * This function does little more than loop over the pipe and call
932 * @actor to do the actual moving of a single struct pipe_buffer to
933 * the desired destination. See pipe_to_file, pipe_to_sendpage, or
934 * pipe_to_user.
935 *
Jens Axboe83f91352006-04-02 23:05:09 +0200936 */
Jens Axboec66ab6f2007-06-12 21:17:17 +0200937ssize_t __splice_from_pipe(struct pipe_inode_info *pipe, struct splice_desc *sd,
938 splice_actor *actor)
Jens Axboe5274f052006-03-30 15:15:30 +0200939{
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200940 int ret;
Jens Axboe5274f052006-03-30 15:15:30 +0200941
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200942 splice_from_pipe_begin(sd);
943 do {
944 ret = splice_from_pipe_next(pipe, sd);
945 if (ret > 0)
946 ret = splice_from_pipe_feed(pipe, sd, actor);
947 } while (ret > 0);
948 splice_from_pipe_end(pipe, sd);
Jens Axboe5274f052006-03-30 15:15:30 +0200949
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200950 return sd->num_spliced ? sd->num_spliced : ret;
Jens Axboe5274f052006-03-30 15:15:30 +0200951}
Mark Fasheh40bee44e2007-03-21 13:11:02 +0100952EXPORT_SYMBOL(__splice_from_pipe);
Jens Axboe5274f052006-03-30 15:15:30 +0200953
Jens Axboe932cc6d2007-06-21 13:10:21 +0200954/**
955 * splice_from_pipe - splice data from a pipe to a file
956 * @pipe: pipe to splice from
957 * @out: file to splice to
958 * @ppos: position in @out
959 * @len: how many bytes to splice
960 * @flags: splice modifier flags
961 * @actor: handler that splices the data
962 *
963 * Description:
Miklos Szeredi29339702009-04-14 19:48:37 +0200964 * See __splice_from_pipe. This function locks the pipe inode,
Jens Axboe932cc6d2007-06-21 13:10:21 +0200965 * otherwise it's identical to __splice_from_pipe().
966 *
967 */
Mark Fasheh6da61802006-10-17 18:43:07 +0200968ssize_t splice_from_pipe(struct pipe_inode_info *pipe, struct file *out,
969 loff_t *ppos, size_t len, unsigned int flags,
970 splice_actor *actor)
971{
972 ssize_t ret;
Jens Axboec66ab6f2007-06-12 21:17:17 +0200973 struct splice_desc sd = {
974 .total_len = len,
975 .flags = flags,
976 .pos = *ppos,
Jens Axboe6a14b902007-06-14 13:08:55 +0200977 .u.file = out,
Jens Axboec66ab6f2007-06-12 21:17:17 +0200978 };
Mark Fasheh6da61802006-10-17 18:43:07 +0200979
Miklos Szeredi61e0d472009-04-14 19:48:41 +0200980 pipe_lock(pipe);
Jens Axboec66ab6f2007-06-12 21:17:17 +0200981 ret = __splice_from_pipe(pipe, &sd, actor);
Miklos Szeredi61e0d472009-04-14 19:48:41 +0200982 pipe_unlock(pipe);
Mark Fasheh6da61802006-10-17 18:43:07 +0200983
984 return ret;
985}
986
987/**
Jens Axboe83f91352006-04-02 23:05:09 +0200988 * generic_file_splice_write - splice data from a pipe to a file
Ingo Molnar3a326a22006-04-10 15:18:35 +0200989 * @pipe: pipe info
Jens Axboe83f91352006-04-02 23:05:09 +0200990 * @out: file to write to
Jens Axboe932cc6d2007-06-21 13:10:21 +0200991 * @ppos: position in @out
Jens Axboe83f91352006-04-02 23:05:09 +0200992 * @len: number of bytes to splice
993 * @flags: splice modifier flags
994 *
Jens Axboe932cc6d2007-06-21 13:10:21 +0200995 * Description:
996 * Will either move or copy pages (determined by @flags options) from
997 * the given pipe inode to the given file.
Jens Axboe83f91352006-04-02 23:05:09 +0200998 *
999 */
Ingo Molnar3a326a22006-04-10 15:18:35 +02001000ssize_t
1001generic_file_splice_write(struct pipe_inode_info *pipe, struct file *out,
Jens Axboecbb7e572006-04-11 14:57:50 +02001002 loff_t *ppos, size_t len, unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +02001003{
Jens Axboe4f6f0bd2006-04-02 23:04:46 +02001004 struct address_space *mapping = out->f_mapping;
Jens Axboe8c34e2d2006-10-17 19:43:22 +02001005 struct inode *inode = mapping->host;
Miklos Szeredi7f3d4ee2008-05-07 09:22:39 +02001006 struct splice_desc sd = {
1007 .total_len = len,
1008 .flags = flags,
1009 .pos = *ppos,
1010 .u.file = out,
1011 };
Ingo Molnar3a326a22006-04-10 15:18:35 +02001012 ssize_t ret;
Jens Axboe8c34e2d2006-10-17 19:43:22 +02001013
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001014 pipe_lock(pipe);
Miklos Szeredieb443e52009-04-14 19:48:38 +02001015
1016 splice_from_pipe_begin(&sd);
1017 do {
1018 ret = splice_from_pipe_next(pipe, &sd);
1019 if (ret <= 0)
1020 break;
1021
1022 mutex_lock_nested(&inode->i_mutex, I_MUTEX_CHILD);
1023 ret = file_remove_suid(out);
Miklos Szeredi723590e2009-08-15 08:43:22 +02001024 if (!ret) {
Josef Bacikc3b2da32012-03-26 09:59:21 -04001025 ret = file_update_time(out);
1026 if (!ret)
1027 ret = splice_from_pipe_feed(pipe, &sd,
1028 pipe_to_file);
Miklos Szeredi723590e2009-08-15 08:43:22 +02001029 }
Miklos Szeredieb443e52009-04-14 19:48:38 +02001030 mutex_unlock(&inode->i_mutex);
1031 } while (ret > 0);
1032 splice_from_pipe_end(pipe, &sd);
1033
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001034 pipe_unlock(pipe);
Miklos Szeredieb443e52009-04-14 19:48:38 +02001035
1036 if (sd.num_spliced)
1037 ret = sd.num_spliced;
1038
Jens Axboea4514eb2006-04-19 15:57:05 +02001039 if (ret > 0) {
Jan Kara148f9482009-08-17 19:52:36 +02001040 int err;
Jens Axboe17ee4f42007-06-15 13:10:37 +02001041
Jan Kara148f9482009-08-17 19:52:36 +02001042 err = generic_write_sync(out, *ppos, ret);
1043 if (err)
1044 ret = err;
1045 else
1046 *ppos += ret;
Namjae Jeond0e1d662012-12-11 16:00:21 -08001047 balance_dirty_pages_ratelimited(mapping);
Jens Axboe4f6f0bd2006-04-02 23:04:46 +02001048 }
1049
1050 return ret;
Jens Axboe5274f052006-03-30 15:15:30 +02001051}
1052
Jens Axboe059a8f32006-04-02 23:06:05 +02001053EXPORT_SYMBOL(generic_file_splice_write);
1054
Miklos Szeredib2858d72009-05-19 11:37:46 +02001055static int write_pipe_buf(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
1056 struct splice_desc *sd)
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +02001057{
Miklos Szeredib2858d72009-05-19 11:37:46 +02001058 int ret;
1059 void *data;
Al Viro06ae43f2013-03-20 13:19:30 -04001060 loff_t tmp = sd->pos;
Miklos Szeredib2858d72009-05-19 11:37:46 +02001061
Al Virofbb32752014-02-02 21:09:54 -05001062 data = kmap(buf->page);
Al Viro06ae43f2013-03-20 13:19:30 -04001063 ret = __kernel_write(sd->u.file, data + buf->offset, sd->len, &tmp);
Al Virofbb32752014-02-02 21:09:54 -05001064 kunmap(buf->page);
Miklos Szeredib2858d72009-05-19 11:37:46 +02001065
1066 return ret;
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +02001067}
1068
1069static ssize_t default_file_splice_write(struct pipe_inode_info *pipe,
1070 struct file *out, loff_t *ppos,
1071 size_t len, unsigned int flags)
1072{
Miklos Szeredib2858d72009-05-19 11:37:46 +02001073 ssize_t ret;
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +02001074
Miklos Szeredib2858d72009-05-19 11:37:46 +02001075 ret = splice_from_pipe(pipe, out, ppos, len, flags, write_pipe_buf);
1076 if (ret > 0)
1077 *ppos += ret;
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +02001078
Miklos Szeredib2858d72009-05-19 11:37:46 +02001079 return ret;
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +02001080}
1081
Jens Axboe83f91352006-04-02 23:05:09 +02001082/**
1083 * generic_splice_sendpage - splice data from a pipe to a socket
Jens Axboe932cc6d2007-06-21 13:10:21 +02001084 * @pipe: pipe to splice from
Jens Axboe83f91352006-04-02 23:05:09 +02001085 * @out: socket to write to
Jens Axboe932cc6d2007-06-21 13:10:21 +02001086 * @ppos: position in @out
Jens Axboe83f91352006-04-02 23:05:09 +02001087 * @len: number of bytes to splice
1088 * @flags: splice modifier flags
1089 *
Jens Axboe932cc6d2007-06-21 13:10:21 +02001090 * Description:
1091 * Will send @len bytes from the pipe to a network socket. No data copying
1092 * is involved.
Jens Axboe83f91352006-04-02 23:05:09 +02001093 *
1094 */
Ingo Molnar3a326a22006-04-10 15:18:35 +02001095ssize_t generic_splice_sendpage(struct pipe_inode_info *pipe, struct file *out,
Jens Axboecbb7e572006-04-11 14:57:50 +02001096 loff_t *ppos, size_t len, unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +02001097{
Jens Axboe00522fb2006-04-26 14:39:29 +02001098 return splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_sendpage);
Jens Axboe5274f052006-03-30 15:15:30 +02001099}
1100
Jens Axboe059a8f32006-04-02 23:06:05 +02001101EXPORT_SYMBOL(generic_splice_sendpage);
Jeff Garzika0f06782006-03-30 23:06:13 -05001102
Jens Axboe83f91352006-04-02 23:05:09 +02001103/*
1104 * Attempt to initiate a splice from pipe to file.
1105 */
Ingo Molnar3a326a22006-04-10 15:18:35 +02001106static long do_splice_from(struct pipe_inode_info *pipe, struct file *out,
Jens Axboecbb7e572006-04-11 14:57:50 +02001107 loff_t *ppos, size_t len, unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +02001108{
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +02001109 ssize_t (*splice_write)(struct pipe_inode_info *, struct file *,
1110 loff_t *, size_t, unsigned int);
Jens Axboe5274f052006-03-30 15:15:30 +02001111
Al Viro72c2d532013-09-22 16:27:52 -04001112 if (out->f_op->splice_write)
Changli Gaocc56f7d2009-11-04 09:09:52 +01001113 splice_write = out->f_op->splice_write;
1114 else
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +02001115 splice_write = default_file_splice_write;
1116
Al Viro500368f2013-05-23 20:07:11 -04001117 return splice_write(pipe, out, ppos, len, flags);
Jens Axboe5274f052006-03-30 15:15:30 +02001118}
1119
Jens Axboe83f91352006-04-02 23:05:09 +02001120/*
1121 * Attempt to initiate a splice from a file to a pipe.
1122 */
Jens Axboecbb7e572006-04-11 14:57:50 +02001123static long do_splice_to(struct file *in, loff_t *ppos,
1124 struct pipe_inode_info *pipe, size_t len,
1125 unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +02001126{
Miklos Szeredi68181732009-05-07 15:37:36 +02001127 ssize_t (*splice_read)(struct file *, loff_t *,
1128 struct pipe_inode_info *, size_t, unsigned int);
Jens Axboe5274f052006-03-30 15:15:30 +02001129 int ret;
1130
Jens Axboe49570e92006-04-11 13:56:09 +02001131 if (unlikely(!(in->f_mode & FMODE_READ)))
Jens Axboe5274f052006-03-30 15:15:30 +02001132 return -EBADF;
1133
Jens Axboecbb7e572006-04-11 14:57:50 +02001134 ret = rw_verify_area(READ, in, ppos, len);
Jens Axboe5274f052006-03-30 15:15:30 +02001135 if (unlikely(ret < 0))
1136 return ret;
1137
Al Viro72c2d532013-09-22 16:27:52 -04001138 if (in->f_op->splice_read)
Changli Gaocc56f7d2009-11-04 09:09:52 +01001139 splice_read = in->f_op->splice_read;
1140 else
Miklos Szeredi68181732009-05-07 15:37:36 +02001141 splice_read = default_file_splice_read;
1142
1143 return splice_read(in, ppos, pipe, len, flags);
Jens Axboe5274f052006-03-30 15:15:30 +02001144}
1145
Jens Axboe932cc6d2007-06-21 13:10:21 +02001146/**
1147 * splice_direct_to_actor - splices data directly between two non-pipes
1148 * @in: file to splice from
1149 * @sd: actor information on where to splice to
1150 * @actor: handles the data splicing
1151 *
1152 * Description:
1153 * This is a special case helper to splice directly between two
1154 * points, without requiring an explicit pipe. Internally an allocated
Randy Dunlap79685b82007-07-27 08:08:51 +02001155 * pipe is cached in the process, and reused during the lifetime of
Jens Axboe932cc6d2007-06-21 13:10:21 +02001156 * that process.
1157 *
Jens Axboec66ab6f2007-06-12 21:17:17 +02001158 */
1159ssize_t splice_direct_to_actor(struct file *in, struct splice_desc *sd,
1160 splice_direct_actor *actor)
Jens Axboeb92ce552006-04-11 13:52:07 +02001161{
1162 struct pipe_inode_info *pipe;
1163 long ret, bytes;
1164 umode_t i_mode;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001165 size_t len;
1166 int i, flags;
Jens Axboeb92ce552006-04-11 13:52:07 +02001167
1168 /*
1169 * We require the input being a regular file, as we don't want to
1170 * randomly drop data for eg socket -> socket splicing. Use the
1171 * piped splicing for that!
1172 */
Al Viro496ad9a2013-01-23 17:07:38 -05001173 i_mode = file_inode(in)->i_mode;
Jens Axboeb92ce552006-04-11 13:52:07 +02001174 if (unlikely(!S_ISREG(i_mode) && !S_ISBLK(i_mode)))
1175 return -EINVAL;
1176
1177 /*
1178 * neither in nor out is a pipe, setup an internal pipe attached to
1179 * 'out' and transfer the wanted data from 'in' to 'out' through that
1180 */
1181 pipe = current->splice_pipe;
Jens Axboe49570e92006-04-11 13:56:09 +02001182 if (unlikely(!pipe)) {
Al Viro7bee1302013-03-21 11:04:15 -04001183 pipe = alloc_pipe_info();
Jens Axboeb92ce552006-04-11 13:52:07 +02001184 if (!pipe)
1185 return -ENOMEM;
1186
1187 /*
1188 * We don't have an immediate reader, but we'll read the stuff
Jens Axboe00522fb2006-04-26 14:39:29 +02001189 * out of the pipe right after the splice_to_pipe(). So set
Jens Axboeb92ce552006-04-11 13:52:07 +02001190 * PIPE_READERS appropriately.
1191 */
1192 pipe->readers = 1;
1193
1194 current->splice_pipe = pipe;
1195 }
1196
1197 /*
Ingo Molnar73d62d82006-04-11 13:57:21 +02001198 * Do the splice.
Jens Axboeb92ce552006-04-11 13:52:07 +02001199 */
1200 ret = 0;
1201 bytes = 0;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001202 len = sd->total_len;
1203 flags = sd->flags;
1204
1205 /*
1206 * Don't block on output, we have to drain the direct pipe.
1207 */
1208 sd->flags &= ~SPLICE_F_NONBLOCK;
Jens Axboeb92ce552006-04-11 13:52:07 +02001209
1210 while (len) {
Jens Axboe51a92c02007-07-13 14:11:43 +02001211 size_t read_len;
Tom Zanussia82c53a2008-05-09 13:28:36 +02001212 loff_t pos = sd->pos, prev_pos = pos;
Jens Axboeb92ce552006-04-11 13:52:07 +02001213
Jens Axboebcd4f3a2007-07-16 14:41:49 +02001214 ret = do_splice_to(in, &pos, pipe, len, flags);
Jens Axboe51a92c02007-07-13 14:11:43 +02001215 if (unlikely(ret <= 0))
Jens Axboeb92ce552006-04-11 13:52:07 +02001216 goto out_release;
1217
1218 read_len = ret;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001219 sd->total_len = read_len;
Jens Axboeb92ce552006-04-11 13:52:07 +02001220
1221 /*
1222 * NOTE: nonblocking mode only applies to the input. We
1223 * must not do the output in nonblocking mode as then we
1224 * could get stuck data in the internal pipe:
1225 */
Jens Axboec66ab6f2007-06-12 21:17:17 +02001226 ret = actor(pipe, sd);
Tom Zanussia82c53a2008-05-09 13:28:36 +02001227 if (unlikely(ret <= 0)) {
1228 sd->pos = prev_pos;
Jens Axboeb92ce552006-04-11 13:52:07 +02001229 goto out_release;
Tom Zanussia82c53a2008-05-09 13:28:36 +02001230 }
Jens Axboeb92ce552006-04-11 13:52:07 +02001231
1232 bytes += ret;
1233 len -= ret;
Jens Axboebcd4f3a2007-07-16 14:41:49 +02001234 sd->pos = pos;
Jens Axboeb92ce552006-04-11 13:52:07 +02001235
Tom Zanussia82c53a2008-05-09 13:28:36 +02001236 if (ret < read_len) {
1237 sd->pos = prev_pos + ret;
Jens Axboe51a92c02007-07-13 14:11:43 +02001238 goto out_release;
Tom Zanussia82c53a2008-05-09 13:28:36 +02001239 }
Jens Axboeb92ce552006-04-11 13:52:07 +02001240 }
1241
Jens Axboe9e971982008-01-29 21:05:57 +01001242done:
Jens Axboeb92ce552006-04-11 13:52:07 +02001243 pipe->nrbufs = pipe->curbuf = 0;
Jens Axboe80848702008-01-30 12:24:48 +01001244 file_accessed(in);
Jens Axboeb92ce552006-04-11 13:52:07 +02001245 return bytes;
1246
1247out_release:
1248 /*
1249 * If we did an incomplete transfer we must release
1250 * the pipe buffers in question:
1251 */
Jens Axboe35f3d142010-05-20 10:43:18 +02001252 for (i = 0; i < pipe->buffers; i++) {
Jens Axboeb92ce552006-04-11 13:52:07 +02001253 struct pipe_buffer *buf = pipe->bufs + i;
1254
1255 if (buf->ops) {
1256 buf->ops->release(pipe, buf);
1257 buf->ops = NULL;
1258 }
1259 }
Jens Axboeb92ce552006-04-11 13:52:07 +02001260
Jens Axboe9e971982008-01-29 21:05:57 +01001261 if (!bytes)
1262 bytes = ret;
Jens Axboeb92ce552006-04-11 13:52:07 +02001263
Jens Axboe9e971982008-01-29 21:05:57 +01001264 goto done;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001265}
1266EXPORT_SYMBOL(splice_direct_to_actor);
1267
1268static int direct_splice_actor(struct pipe_inode_info *pipe,
1269 struct splice_desc *sd)
1270{
Jens Axboe6a14b902007-06-14 13:08:55 +02001271 struct file *file = sd->u.file;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001272
Al Viro7995bd22013-06-20 18:58:36 +04001273 return do_splice_from(pipe, file, sd->opos, sd->total_len,
Changli Gao2cb4b052010-06-29 13:09:18 +02001274 sd->flags);
Jens Axboec66ab6f2007-06-12 21:17:17 +02001275}
1276
Jens Axboe932cc6d2007-06-21 13:10:21 +02001277/**
1278 * do_splice_direct - splices data directly between two files
1279 * @in: file to splice from
1280 * @ppos: input file offset
1281 * @out: file to splice to
Randy Dunlapacdb37c2013-06-22 19:44:08 -07001282 * @opos: output file offset
Jens Axboe932cc6d2007-06-21 13:10:21 +02001283 * @len: number of bytes to splice
1284 * @flags: splice modifier flags
1285 *
1286 * Description:
1287 * For use by do_sendfile(). splice can easily emulate sendfile, but
1288 * doing it in the application would incur an extra system call
1289 * (splice in + splice out, as compared to just sendfile()). So this helper
1290 * can splice directly through a process-private pipe.
1291 *
1292 */
Jens Axboec66ab6f2007-06-12 21:17:17 +02001293long do_splice_direct(struct file *in, loff_t *ppos, struct file *out,
Al Viro7995bd22013-06-20 18:58:36 +04001294 loff_t *opos, size_t len, unsigned int flags)
Jens Axboec66ab6f2007-06-12 21:17:17 +02001295{
1296 struct splice_desc sd = {
1297 .len = len,
1298 .total_len = len,
1299 .flags = flags,
1300 .pos = *ppos,
Jens Axboe6a14b902007-06-14 13:08:55 +02001301 .u.file = out,
Al Viro7995bd22013-06-20 18:58:36 +04001302 .opos = opos,
Jens Axboec66ab6f2007-06-12 21:17:17 +02001303 };
Jens Axboe51a92c02007-07-13 14:11:43 +02001304 long ret;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001305
Al Viro18c67cb2013-06-19 15:41:54 +04001306 if (unlikely(!(out->f_mode & FMODE_WRITE)))
1307 return -EBADF;
1308
1309 if (unlikely(out->f_flags & O_APPEND))
1310 return -EINVAL;
1311
1312 ret = rw_verify_area(WRITE, out, opos, len);
1313 if (unlikely(ret < 0))
1314 return ret;
1315
Jens Axboec66ab6f2007-06-12 21:17:17 +02001316 ret = splice_direct_to_actor(in, &sd, direct_splice_actor);
Jens Axboe51a92c02007-07-13 14:11:43 +02001317 if (ret > 0)
Tom Zanussia82c53a2008-05-09 13:28:36 +02001318 *ppos = sd.pos;
Jens Axboe51a92c02007-07-13 14:11:43 +02001319
Jens Axboec66ab6f2007-06-12 21:17:17 +02001320 return ret;
Jens Axboeb92ce552006-04-11 13:52:07 +02001321}
1322
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001323static int splice_pipe_to_pipe(struct pipe_inode_info *ipipe,
1324 struct pipe_inode_info *opipe,
1325 size_t len, unsigned int flags);
Jens Axboeddac0d32006-11-04 12:49:32 +01001326
1327/*
Jens Axboe83f91352006-04-02 23:05:09 +02001328 * Determine where to splice to/from.
1329 */
Ingo Molnar529565d2006-04-10 15:18:58 +02001330static long do_splice(struct file *in, loff_t __user *off_in,
1331 struct file *out, loff_t __user *off_out,
1332 size_t len, unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +02001333{
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001334 struct pipe_inode_info *ipipe;
1335 struct pipe_inode_info *opipe;
Al Viro7995bd22013-06-20 18:58:36 +04001336 loff_t offset;
Jens Axboea4514eb2006-04-19 15:57:05 +02001337 long ret;
Jens Axboe5274f052006-03-30 15:15:30 +02001338
Linus Torvalds71993e62010-11-28 13:56:09 -08001339 ipipe = get_pipe_info(in);
1340 opipe = get_pipe_info(out);
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001341
1342 if (ipipe && opipe) {
1343 if (off_in || off_out)
1344 return -ESPIPE;
1345
1346 if (!(in->f_mode & FMODE_READ))
1347 return -EBADF;
1348
1349 if (!(out->f_mode & FMODE_WRITE))
1350 return -EBADF;
1351
1352 /* Splicing to self would be fun, but... */
1353 if (ipipe == opipe)
1354 return -EINVAL;
1355
1356 return splice_pipe_to_pipe(ipipe, opipe, len, flags);
1357 }
1358
1359 if (ipipe) {
Ingo Molnar529565d2006-04-10 15:18:58 +02001360 if (off_in)
1361 return -ESPIPE;
Jens Axboeb92ce552006-04-11 13:52:07 +02001362 if (off_out) {
Changli Gao19c9a492010-06-29 13:10:36 +02001363 if (!(out->f_mode & FMODE_PWRITE))
Jens Axboeb92ce552006-04-11 13:52:07 +02001364 return -EINVAL;
Jens Axboecbb7e572006-04-11 14:57:50 +02001365 if (copy_from_user(&offset, off_out, sizeof(loff_t)))
Jens Axboeb92ce552006-04-11 13:52:07 +02001366 return -EFAULT;
Al Viro7995bd22013-06-20 18:58:36 +04001367 } else {
1368 offset = out->f_pos;
1369 }
Ingo Molnar529565d2006-04-10 15:18:58 +02001370
Al Viro18c67cb2013-06-19 15:41:54 +04001371 if (unlikely(!(out->f_mode & FMODE_WRITE)))
1372 return -EBADF;
1373
1374 if (unlikely(out->f_flags & O_APPEND))
1375 return -EINVAL;
1376
1377 ret = rw_verify_area(WRITE, out, &offset, len);
1378 if (unlikely(ret < 0))
1379 return ret;
1380
Al Viro500368f2013-05-23 20:07:11 -04001381 file_start_write(out);
Al Viro7995bd22013-06-20 18:58:36 +04001382 ret = do_splice_from(ipipe, out, &offset, len, flags);
Al Viro500368f2013-05-23 20:07:11 -04001383 file_end_write(out);
Jens Axboea4514eb2006-04-19 15:57:05 +02001384
Al Viro7995bd22013-06-20 18:58:36 +04001385 if (!off_out)
1386 out->f_pos = offset;
1387 else if (copy_to_user(off_out, &offset, sizeof(loff_t)))
Jens Axboea4514eb2006-04-19 15:57:05 +02001388 ret = -EFAULT;
1389
1390 return ret;
Ingo Molnar529565d2006-04-10 15:18:58 +02001391 }
Jens Axboe5274f052006-03-30 15:15:30 +02001392
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001393 if (opipe) {
Ingo Molnar529565d2006-04-10 15:18:58 +02001394 if (off_out)
1395 return -ESPIPE;
Jens Axboeb92ce552006-04-11 13:52:07 +02001396 if (off_in) {
Changli Gao19c9a492010-06-29 13:10:36 +02001397 if (!(in->f_mode & FMODE_PREAD))
Jens Axboeb92ce552006-04-11 13:52:07 +02001398 return -EINVAL;
Jens Axboecbb7e572006-04-11 14:57:50 +02001399 if (copy_from_user(&offset, off_in, sizeof(loff_t)))
Jens Axboeb92ce552006-04-11 13:52:07 +02001400 return -EFAULT;
Al Viro7995bd22013-06-20 18:58:36 +04001401 } else {
1402 offset = in->f_pos;
1403 }
Ingo Molnar529565d2006-04-10 15:18:58 +02001404
Al Viro7995bd22013-06-20 18:58:36 +04001405 ret = do_splice_to(in, &offset, opipe, len, flags);
Jens Axboea4514eb2006-04-19 15:57:05 +02001406
Al Viro7995bd22013-06-20 18:58:36 +04001407 if (!off_in)
1408 in->f_pos = offset;
1409 else if (copy_to_user(off_in, &offset, sizeof(loff_t)))
Jens Axboea4514eb2006-04-19 15:57:05 +02001410 ret = -EFAULT;
1411
1412 return ret;
Ingo Molnar529565d2006-04-10 15:18:58 +02001413 }
Jens Axboe5274f052006-03-30 15:15:30 +02001414
1415 return -EINVAL;
1416}
1417
Jens Axboe912d35f2006-04-26 10:59:21 +02001418/*
1419 * Map an iov into an array of pages and offset/length tupples. With the
1420 * partial_page structure, we can map several non-contiguous ranges into
1421 * our ones pages[] map instead of splitting that operation into pieces.
1422 * Could easily be exported as a generic helper for other users, in which
1423 * case one would probably want to add a 'max_nr_pages' parameter as well.
1424 */
1425static int get_iovec_page_array(const struct iovec __user *iov,
1426 unsigned int nr_vecs, struct page **pages,
Eric Dumazetbd1a68b2012-04-04 11:08:51 +02001427 struct partial_page *partial, bool aligned,
Jens Axboe35f3d142010-05-20 10:43:18 +02001428 unsigned int pipe_buffers)
Jens Axboe912d35f2006-04-26 10:59:21 +02001429{
1430 int buffers = 0, error = 0;
1431
Jens Axboe912d35f2006-04-26 10:59:21 +02001432 while (nr_vecs) {
1433 unsigned long off, npages;
Linus Torvalds75723952007-10-01 13:17:28 -07001434 struct iovec entry;
Jens Axboe912d35f2006-04-26 10:59:21 +02001435 void __user *base;
1436 size_t len;
1437 int i;
1438
Linus Torvalds75723952007-10-01 13:17:28 -07001439 error = -EFAULT;
Nick Pigginbc40d732008-07-25 19:45:26 -07001440 if (copy_from_user(&entry, iov, sizeof(entry)))
Jens Axboe912d35f2006-04-26 10:59:21 +02001441 break;
Linus Torvalds75723952007-10-01 13:17:28 -07001442
1443 base = entry.iov_base;
1444 len = entry.iov_len;
Jens Axboe912d35f2006-04-26 10:59:21 +02001445
1446 /*
1447 * Sanity check this iovec. 0 read succeeds.
1448 */
Linus Torvalds75723952007-10-01 13:17:28 -07001449 error = 0;
Jens Axboe912d35f2006-04-26 10:59:21 +02001450 if (unlikely(!len))
1451 break;
1452 error = -EFAULT;
Bastian Blank712a30e2008-02-10 16:47:57 +02001453 if (!access_ok(VERIFY_READ, base, len))
Jens Axboe912d35f2006-04-26 10:59:21 +02001454 break;
1455
1456 /*
1457 * Get this base offset and number of pages, then map
1458 * in the user pages.
1459 */
1460 off = (unsigned long) base & ~PAGE_MASK;
Jens Axboe7afa6fd2006-05-01 20:02:33 +02001461
1462 /*
1463 * If asked for alignment, the offset must be zero and the
1464 * length a multiple of the PAGE_SIZE.
1465 */
1466 error = -EINVAL;
1467 if (aligned && (off || len & ~PAGE_MASK))
1468 break;
1469
Jens Axboe912d35f2006-04-26 10:59:21 +02001470 npages = (off + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
Jens Axboe35f3d142010-05-20 10:43:18 +02001471 if (npages > pipe_buffers - buffers)
1472 npages = pipe_buffers - buffers;
Jens Axboe912d35f2006-04-26 10:59:21 +02001473
Nick Pigginbc40d732008-07-25 19:45:26 -07001474 error = get_user_pages_fast((unsigned long)base, npages,
1475 0, &pages[buffers]);
Jens Axboe912d35f2006-04-26 10:59:21 +02001476
1477 if (unlikely(error <= 0))
1478 break;
1479
1480 /*
1481 * Fill this contiguous range into the partial page map.
1482 */
1483 for (i = 0; i < error; i++) {
Jens Axboe75914892006-05-02 12:57:18 +02001484 const int plen = min_t(size_t, len, PAGE_SIZE - off);
Jens Axboe912d35f2006-04-26 10:59:21 +02001485
1486 partial[buffers].offset = off;
1487 partial[buffers].len = plen;
1488
1489 off = 0;
1490 len -= plen;
1491 buffers++;
1492 }
1493
1494 /*
1495 * We didn't complete this iov, stop here since it probably
1496 * means we have to move some of this into a pipe to
1497 * be able to continue.
1498 */
1499 if (len)
1500 break;
1501
1502 /*
1503 * Don't continue if we mapped fewer pages than we asked for,
1504 * or if we mapped the max number of pages that we have
1505 * room for.
1506 */
Jens Axboe35f3d142010-05-20 10:43:18 +02001507 if (error < npages || buffers == pipe_buffers)
Jens Axboe912d35f2006-04-26 10:59:21 +02001508 break;
1509
1510 nr_vecs--;
1511 iov++;
1512 }
1513
Jens Axboe912d35f2006-04-26 10:59:21 +02001514 if (buffers)
1515 return buffers;
1516
1517 return error;
1518}
1519
Jens Axboe6a14b902007-06-14 13:08:55 +02001520static int pipe_to_user(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
1521 struct splice_desc *sd)
1522{
Al Viro6130f532014-02-03 18:19:51 -05001523 int n = copy_page_to_iter(buf->page, buf->offset, sd->len, sd->u.data);
1524 return n == sd->len ? n : -EFAULT;
Jens Axboe6a14b902007-06-14 13:08:55 +02001525}
1526
1527/*
1528 * For lack of a better implementation, implement vmsplice() to userspace
1529 * as a simple copy of the pipes pages to the user iov.
1530 */
Al Viro6130f532014-02-03 18:19:51 -05001531static long vmsplice_to_user(struct file *file, const struct iovec __user *uiov,
Jens Axboe6a14b902007-06-14 13:08:55 +02001532 unsigned long nr_segs, unsigned int flags)
1533{
1534 struct pipe_inode_info *pipe;
1535 struct splice_desc sd;
Jens Axboe6a14b902007-06-14 13:08:55 +02001536 long ret;
Al Viro6130f532014-02-03 18:19:51 -05001537 struct iovec iovstack[UIO_FASTIOV];
1538 struct iovec *iov = iovstack;
1539 struct iov_iter iter;
Miklos Szeredib6dd6f42014-05-27 16:41:16 +02001540 ssize_t count;
Jens Axboe6a14b902007-06-14 13:08:55 +02001541
Linus Torvalds71993e62010-11-28 13:56:09 -08001542 pipe = get_pipe_info(file);
Jens Axboe6a14b902007-06-14 13:08:55 +02001543 if (!pipe)
1544 return -EBADF;
1545
Al Viro6130f532014-02-03 18:19:51 -05001546 ret = rw_copy_check_uvector(READ, uiov, nr_segs,
1547 ARRAY_SIZE(iovstack), iovstack, &iov);
1548 if (ret <= 0)
Miklos Szeredib6dd6f42014-05-27 16:41:16 +02001549 goto out;
Al Viro6130f532014-02-03 18:19:51 -05001550
Miklos Szeredib6dd6f42014-05-27 16:41:16 +02001551 count = ret;
Al Viro6130f532014-02-03 18:19:51 -05001552 iov_iter_init(&iter, iov, nr_segs, count, 0);
1553
1554 sd.len = 0;
1555 sd.total_len = count;
1556 sd.flags = flags;
1557 sd.u.data = &iter;
1558 sd.pos = 0;
1559
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001560 pipe_lock(pipe);
Al Viro6130f532014-02-03 18:19:51 -05001561 ret = __splice_from_pipe(pipe, &sd, pipe_to_user);
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001562 pipe_unlock(pipe);
Jens Axboe6a14b902007-06-14 13:08:55 +02001563
Miklos Szeredib6dd6f42014-05-27 16:41:16 +02001564out:
Al Viro6130f532014-02-03 18:19:51 -05001565 if (iov != iovstack)
1566 kfree(iov);
Jens Axboe6a14b902007-06-14 13:08:55 +02001567
1568 return ret;
1569}
1570
Jens Axboe912d35f2006-04-26 10:59:21 +02001571/*
1572 * vmsplice splices a user address range into a pipe. It can be thought of
1573 * as splice-from-memory, where the regular splice is splice-from-file (or
1574 * to file). In both cases the output is a pipe, naturally.
Jens Axboe912d35f2006-04-26 10:59:21 +02001575 */
Jens Axboe6a14b902007-06-14 13:08:55 +02001576static long vmsplice_to_pipe(struct file *file, const struct iovec __user *iov,
1577 unsigned long nr_segs, unsigned int flags)
Jens Axboe912d35f2006-04-26 10:59:21 +02001578{
Jens Axboeddac0d32006-11-04 12:49:32 +01001579 struct pipe_inode_info *pipe;
Jens Axboe35f3d142010-05-20 10:43:18 +02001580 struct page *pages[PIPE_DEF_BUFFERS];
1581 struct partial_page partial[PIPE_DEF_BUFFERS];
Jens Axboe912d35f2006-04-26 10:59:21 +02001582 struct splice_pipe_desc spd = {
1583 .pages = pages,
1584 .partial = partial,
Eric Dumazet047fe362012-06-12 15:24:40 +02001585 .nr_pages_max = PIPE_DEF_BUFFERS,
Jens Axboe912d35f2006-04-26 10:59:21 +02001586 .flags = flags,
1587 .ops = &user_page_pipe_buf_ops,
Jens Axboebbdfc2f2007-11-06 23:29:47 -08001588 .spd_release = spd_release_page,
Jens Axboe912d35f2006-04-26 10:59:21 +02001589 };
Jens Axboe35f3d142010-05-20 10:43:18 +02001590 long ret;
Jens Axboe912d35f2006-04-26 10:59:21 +02001591
Linus Torvalds71993e62010-11-28 13:56:09 -08001592 pipe = get_pipe_info(file);
Jens Axboeddac0d32006-11-04 12:49:32 +01001593 if (!pipe)
Jens Axboe912d35f2006-04-26 10:59:21 +02001594 return -EBADF;
Jens Axboe912d35f2006-04-26 10:59:21 +02001595
Jens Axboe35f3d142010-05-20 10:43:18 +02001596 if (splice_grow_spd(pipe, &spd))
1597 return -ENOMEM;
Jens Axboe912d35f2006-04-26 10:59:21 +02001598
Jens Axboe35f3d142010-05-20 10:43:18 +02001599 spd.nr_pages = get_iovec_page_array(iov, nr_segs, spd.pages,
Eric Dumazetbd1a68b2012-04-04 11:08:51 +02001600 spd.partial, false,
Eric Dumazet047fe362012-06-12 15:24:40 +02001601 spd.nr_pages_max);
Jens Axboe35f3d142010-05-20 10:43:18 +02001602 if (spd.nr_pages <= 0)
1603 ret = spd.nr_pages;
1604 else
1605 ret = splice_to_pipe(pipe, &spd);
1606
Eric Dumazet047fe362012-06-12 15:24:40 +02001607 splice_shrink_spd(&spd);
Jens Axboe35f3d142010-05-20 10:43:18 +02001608 return ret;
Jens Axboe912d35f2006-04-26 10:59:21 +02001609}
1610
Jens Axboe6a14b902007-06-14 13:08:55 +02001611/*
1612 * Note that vmsplice only really supports true splicing _from_ user memory
1613 * to a pipe, not the other way around. Splicing from user memory is a simple
1614 * operation that can be supported without any funky alignment restrictions
1615 * or nasty vm tricks. We simply map in the user memory and fill them into
1616 * a pipe. The reverse isn't quite as easy, though. There are two possible
1617 * solutions for that:
1618 *
1619 * - memcpy() the data internally, at which point we might as well just
1620 * do a regular read() on the buffer anyway.
1621 * - Lots of nasty vm tricks, that are neither fast nor flexible (it
1622 * has restriction limitations on both ends of the pipe).
1623 *
1624 * Currently we punt and implement it as a normal copy, see pipe_to_user().
1625 *
1626 */
Heiko Carstens836f92a2009-01-14 14:14:33 +01001627SYSCALL_DEFINE4(vmsplice, int, fd, const struct iovec __user *, iov,
1628 unsigned long, nr_segs, unsigned int, flags)
Jens Axboe912d35f2006-04-26 10:59:21 +02001629{
Al Viro2903ff02012-08-28 12:52:22 -04001630 struct fd f;
Jens Axboe912d35f2006-04-26 10:59:21 +02001631 long error;
Jens Axboe912d35f2006-04-26 10:59:21 +02001632
Jens Axboe6a14b902007-06-14 13:08:55 +02001633 if (unlikely(nr_segs > UIO_MAXIOV))
1634 return -EINVAL;
1635 else if (unlikely(!nr_segs))
1636 return 0;
1637
Jens Axboe912d35f2006-04-26 10:59:21 +02001638 error = -EBADF;
Al Viro2903ff02012-08-28 12:52:22 -04001639 f = fdget(fd);
1640 if (f.file) {
1641 if (f.file->f_mode & FMODE_WRITE)
1642 error = vmsplice_to_pipe(f.file, iov, nr_segs, flags);
1643 else if (f.file->f_mode & FMODE_READ)
1644 error = vmsplice_to_user(f.file, iov, nr_segs, flags);
Jens Axboe912d35f2006-04-26 10:59:21 +02001645
Al Viro2903ff02012-08-28 12:52:22 -04001646 fdput(f);
Jens Axboe912d35f2006-04-26 10:59:21 +02001647 }
1648
1649 return error;
1650}
1651
Al Viro76b021d2013-03-02 10:19:56 -05001652#ifdef CONFIG_COMPAT
1653COMPAT_SYSCALL_DEFINE4(vmsplice, int, fd, const struct compat_iovec __user *, iov32,
1654 unsigned int, nr_segs, unsigned int, flags)
1655{
1656 unsigned i;
1657 struct iovec __user *iov;
1658 if (nr_segs > UIO_MAXIOV)
1659 return -EINVAL;
1660 iov = compat_alloc_user_space(nr_segs * sizeof(struct iovec));
1661 for (i = 0; i < nr_segs; i++) {
1662 struct compat_iovec v;
1663 if (get_user(v.iov_base, &iov32[i].iov_base) ||
1664 get_user(v.iov_len, &iov32[i].iov_len) ||
1665 put_user(compat_ptr(v.iov_base), &iov[i].iov_base) ||
1666 put_user(v.iov_len, &iov[i].iov_len))
1667 return -EFAULT;
1668 }
1669 return sys_vmsplice(fd, iov, nr_segs, flags);
1670}
1671#endif
1672
Heiko Carstens836f92a2009-01-14 14:14:33 +01001673SYSCALL_DEFINE6(splice, int, fd_in, loff_t __user *, off_in,
1674 int, fd_out, loff_t __user *, off_out,
1675 size_t, len, unsigned int, flags)
Jens Axboe5274f052006-03-30 15:15:30 +02001676{
Al Viro2903ff02012-08-28 12:52:22 -04001677 struct fd in, out;
Jens Axboe5274f052006-03-30 15:15:30 +02001678 long error;
Jens Axboe5274f052006-03-30 15:15:30 +02001679
1680 if (unlikely(!len))
1681 return 0;
1682
1683 error = -EBADF;
Al Viro2903ff02012-08-28 12:52:22 -04001684 in = fdget(fd_in);
1685 if (in.file) {
1686 if (in.file->f_mode & FMODE_READ) {
1687 out = fdget(fd_out);
1688 if (out.file) {
1689 if (out.file->f_mode & FMODE_WRITE)
1690 error = do_splice(in.file, off_in,
1691 out.file, off_out,
Ingo Molnar529565d2006-04-10 15:18:58 +02001692 len, flags);
Al Viro2903ff02012-08-28 12:52:22 -04001693 fdput(out);
Jens Axboe5274f052006-03-30 15:15:30 +02001694 }
1695 }
Al Viro2903ff02012-08-28 12:52:22 -04001696 fdput(in);
Jens Axboe5274f052006-03-30 15:15:30 +02001697 }
Jens Axboe5274f052006-03-30 15:15:30 +02001698 return error;
1699}
Jens Axboe70524492006-04-11 15:51:17 +02001700
1701/*
Jens Axboeaadd06e2006-07-10 11:00:01 +02001702 * Make sure there's data to read. Wait for input if we can, otherwise
1703 * return an appropriate error.
1704 */
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001705static int ipipe_prep(struct pipe_inode_info *pipe, unsigned int flags)
Jens Axboeaadd06e2006-07-10 11:00:01 +02001706{
1707 int ret;
1708
1709 /*
1710 * Check ->nrbufs without the inode lock first. This function
1711 * is speculative anyways, so missing one is ok.
1712 */
1713 if (pipe->nrbufs)
1714 return 0;
1715
1716 ret = 0;
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001717 pipe_lock(pipe);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001718
1719 while (!pipe->nrbufs) {
1720 if (signal_pending(current)) {
1721 ret = -ERESTARTSYS;
1722 break;
1723 }
1724 if (!pipe->writers)
1725 break;
1726 if (!pipe->waiting_writers) {
1727 if (flags & SPLICE_F_NONBLOCK) {
1728 ret = -EAGAIN;
1729 break;
1730 }
1731 }
1732 pipe_wait(pipe);
1733 }
1734
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001735 pipe_unlock(pipe);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001736 return ret;
1737}
1738
1739/*
1740 * Make sure there's writeable room. Wait for room if we can, otherwise
1741 * return an appropriate error.
1742 */
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001743static int opipe_prep(struct pipe_inode_info *pipe, unsigned int flags)
Jens Axboeaadd06e2006-07-10 11:00:01 +02001744{
1745 int ret;
1746
1747 /*
1748 * Check ->nrbufs without the inode lock first. This function
1749 * is speculative anyways, so missing one is ok.
1750 */
Jens Axboe35f3d142010-05-20 10:43:18 +02001751 if (pipe->nrbufs < pipe->buffers)
Jens Axboeaadd06e2006-07-10 11:00:01 +02001752 return 0;
1753
1754 ret = 0;
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001755 pipe_lock(pipe);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001756
Jens Axboe35f3d142010-05-20 10:43:18 +02001757 while (pipe->nrbufs >= pipe->buffers) {
Jens Axboeaadd06e2006-07-10 11:00:01 +02001758 if (!pipe->readers) {
1759 send_sig(SIGPIPE, current, 0);
1760 ret = -EPIPE;
1761 break;
1762 }
1763 if (flags & SPLICE_F_NONBLOCK) {
1764 ret = -EAGAIN;
1765 break;
1766 }
1767 if (signal_pending(current)) {
1768 ret = -ERESTARTSYS;
1769 break;
1770 }
1771 pipe->waiting_writers++;
1772 pipe_wait(pipe);
1773 pipe->waiting_writers--;
1774 }
1775
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001776 pipe_unlock(pipe);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001777 return ret;
1778}
1779
1780/*
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001781 * Splice contents of ipipe to opipe.
1782 */
1783static int splice_pipe_to_pipe(struct pipe_inode_info *ipipe,
1784 struct pipe_inode_info *opipe,
1785 size_t len, unsigned int flags)
1786{
1787 struct pipe_buffer *ibuf, *obuf;
1788 int ret = 0, nbuf;
1789 bool input_wakeup = false;
1790
1791
1792retry:
1793 ret = ipipe_prep(ipipe, flags);
1794 if (ret)
1795 return ret;
1796
1797 ret = opipe_prep(opipe, flags);
1798 if (ret)
1799 return ret;
1800
1801 /*
1802 * Potential ABBA deadlock, work around it by ordering lock
1803 * grabbing by pipe info address. Otherwise two different processes
1804 * could deadlock (one doing tee from A -> B, the other from B -> A).
1805 */
1806 pipe_double_lock(ipipe, opipe);
1807
1808 do {
1809 if (!opipe->readers) {
1810 send_sig(SIGPIPE, current, 0);
1811 if (!ret)
1812 ret = -EPIPE;
1813 break;
1814 }
1815
1816 if (!ipipe->nrbufs && !ipipe->writers)
1817 break;
1818
1819 /*
1820 * Cannot make any progress, because either the input
1821 * pipe is empty or the output pipe is full.
1822 */
Jens Axboe35f3d142010-05-20 10:43:18 +02001823 if (!ipipe->nrbufs || opipe->nrbufs >= opipe->buffers) {
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001824 /* Already processed some buffers, break */
1825 if (ret)
1826 break;
1827
1828 if (flags & SPLICE_F_NONBLOCK) {
1829 ret = -EAGAIN;
1830 break;
1831 }
1832
1833 /*
1834 * We raced with another reader/writer and haven't
1835 * managed to process any buffers. A zero return
1836 * value means EOF, so retry instead.
1837 */
1838 pipe_unlock(ipipe);
1839 pipe_unlock(opipe);
1840 goto retry;
1841 }
1842
1843 ibuf = ipipe->bufs + ipipe->curbuf;
Jens Axboe35f3d142010-05-20 10:43:18 +02001844 nbuf = (opipe->curbuf + opipe->nrbufs) & (opipe->buffers - 1);
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001845 obuf = opipe->bufs + nbuf;
1846
1847 if (len >= ibuf->len) {
1848 /*
1849 * Simply move the whole buffer from ipipe to opipe
1850 */
1851 *obuf = *ibuf;
1852 ibuf->ops = NULL;
1853 opipe->nrbufs++;
Jens Axboe35f3d142010-05-20 10:43:18 +02001854 ipipe->curbuf = (ipipe->curbuf + 1) & (ipipe->buffers - 1);
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001855 ipipe->nrbufs--;
1856 input_wakeup = true;
1857 } else {
1858 /*
1859 * Get a reference to this pipe buffer,
1860 * so we can copy the contents over.
1861 */
1862 ibuf->ops->get(ipipe, ibuf);
1863 *obuf = *ibuf;
1864
1865 /*
1866 * Don't inherit the gift flag, we need to
1867 * prevent multiple steals of this page.
1868 */
1869 obuf->flags &= ~PIPE_BUF_FLAG_GIFT;
1870
1871 obuf->len = len;
1872 opipe->nrbufs++;
1873 ibuf->offset += obuf->len;
1874 ibuf->len -= obuf->len;
1875 }
1876 ret += obuf->len;
1877 len -= obuf->len;
1878 } while (len);
1879
1880 pipe_unlock(ipipe);
1881 pipe_unlock(opipe);
1882
1883 /*
1884 * If we put data in the output pipe, wakeup any potential readers.
1885 */
Namhyung Kim825cdcb2011-05-23 19:58:53 +02001886 if (ret > 0)
1887 wakeup_pipe_readers(opipe);
1888
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001889 if (input_wakeup)
1890 wakeup_pipe_writers(ipipe);
1891
1892 return ret;
1893}
1894
1895/*
Jens Axboe70524492006-04-11 15:51:17 +02001896 * Link contents of ipipe to opipe.
1897 */
1898static int link_pipe(struct pipe_inode_info *ipipe,
1899 struct pipe_inode_info *opipe,
1900 size_t len, unsigned int flags)
1901{
1902 struct pipe_buffer *ibuf, *obuf;
Jens Axboeaadd06e2006-07-10 11:00:01 +02001903 int ret = 0, i = 0, nbuf;
Jens Axboe70524492006-04-11 15:51:17 +02001904
1905 /*
1906 * Potential ABBA deadlock, work around it by ordering lock
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001907 * grabbing by pipe info address. Otherwise two different processes
Jens Axboe70524492006-04-11 15:51:17 +02001908 * could deadlock (one doing tee from A -> B, the other from B -> A).
1909 */
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001910 pipe_double_lock(ipipe, opipe);
Jens Axboe70524492006-04-11 15:51:17 +02001911
Jens Axboeaadd06e2006-07-10 11:00:01 +02001912 do {
Jens Axboe70524492006-04-11 15:51:17 +02001913 if (!opipe->readers) {
1914 send_sig(SIGPIPE, current, 0);
1915 if (!ret)
1916 ret = -EPIPE;
1917 break;
1918 }
Jens Axboe70524492006-04-11 15:51:17 +02001919
1920 /*
Jens Axboeaadd06e2006-07-10 11:00:01 +02001921 * If we have iterated all input buffers or ran out of
1922 * output room, break.
Jens Axboe70524492006-04-11 15:51:17 +02001923 */
Jens Axboe35f3d142010-05-20 10:43:18 +02001924 if (i >= ipipe->nrbufs || opipe->nrbufs >= opipe->buffers)
Jens Axboe70524492006-04-11 15:51:17 +02001925 break;
Jens Axboeaadd06e2006-07-10 11:00:01 +02001926
Jens Axboe35f3d142010-05-20 10:43:18 +02001927 ibuf = ipipe->bufs + ((ipipe->curbuf + i) & (ipipe->buffers-1));
1928 nbuf = (opipe->curbuf + opipe->nrbufs) & (opipe->buffers - 1);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001929
Jens Axboe2a27250e2006-04-19 15:56:40 +02001930 /*
Jens Axboeaadd06e2006-07-10 11:00:01 +02001931 * Get a reference to this pipe buffer,
1932 * so we can copy the contents over.
Jens Axboe2a27250e2006-04-19 15:56:40 +02001933 */
Jens Axboeaadd06e2006-07-10 11:00:01 +02001934 ibuf->ops->get(ipipe, ibuf);
Jens Axboe70524492006-04-11 15:51:17 +02001935
Jens Axboeaadd06e2006-07-10 11:00:01 +02001936 obuf = opipe->bufs + nbuf;
1937 *obuf = *ibuf;
Jens Axboe70524492006-04-11 15:51:17 +02001938
Jens Axboeaadd06e2006-07-10 11:00:01 +02001939 /*
1940 * Don't inherit the gift flag, we need to
1941 * prevent multiple steals of this page.
1942 */
1943 obuf->flags &= ~PIPE_BUF_FLAG_GIFT;
1944
1945 if (obuf->len > len)
1946 obuf->len = len;
1947
1948 opipe->nrbufs++;
1949 ret += obuf->len;
1950 len -= obuf->len;
1951 i++;
1952 } while (len);
Jens Axboe70524492006-04-11 15:51:17 +02001953
Jens Axboe02cf01a2008-02-20 10:34:51 +01001954 /*
1955 * return EAGAIN if we have the potential of some data in the
1956 * future, otherwise just return 0
1957 */
1958 if (!ret && ipipe->waiting_writers && (flags & SPLICE_F_NONBLOCK))
1959 ret = -EAGAIN;
1960
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001961 pipe_unlock(ipipe);
1962 pipe_unlock(opipe);
Jens Axboe70524492006-04-11 15:51:17 +02001963
Jens Axboeaadd06e2006-07-10 11:00:01 +02001964 /*
1965 * If we put data in the output pipe, wakeup any potential readers.
1966 */
Namhyung Kim825cdcb2011-05-23 19:58:53 +02001967 if (ret > 0)
1968 wakeup_pipe_readers(opipe);
Jens Axboe70524492006-04-11 15:51:17 +02001969
1970 return ret;
1971}
1972
1973/*
1974 * This is a tee(1) implementation that works on pipes. It doesn't copy
1975 * any data, it simply references the 'in' pages on the 'out' pipe.
1976 * The 'flags' used are the SPLICE_F_* variants, currently the only
1977 * applicable one is SPLICE_F_NONBLOCK.
1978 */
1979static long do_tee(struct file *in, struct file *out, size_t len,
1980 unsigned int flags)
1981{
Linus Torvalds71993e62010-11-28 13:56:09 -08001982 struct pipe_inode_info *ipipe = get_pipe_info(in);
1983 struct pipe_inode_info *opipe = get_pipe_info(out);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001984 int ret = -EINVAL;
Jens Axboe70524492006-04-11 15:51:17 +02001985
1986 /*
Jens Axboeaadd06e2006-07-10 11:00:01 +02001987 * Duplicate the contents of ipipe to opipe without actually
1988 * copying the data.
Jens Axboe70524492006-04-11 15:51:17 +02001989 */
Jens Axboeaadd06e2006-07-10 11:00:01 +02001990 if (ipipe && opipe && ipipe != opipe) {
1991 /*
1992 * Keep going, unless we encounter an error. The ipipe/opipe
1993 * ordering doesn't really matter.
1994 */
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001995 ret = ipipe_prep(ipipe, flags);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001996 if (!ret) {
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001997 ret = opipe_prep(opipe, flags);
Jens Axboe02cf01a2008-02-20 10:34:51 +01001998 if (!ret)
Jens Axboeaadd06e2006-07-10 11:00:01 +02001999 ret = link_pipe(ipipe, opipe, len, flags);
Jens Axboeaadd06e2006-07-10 11:00:01 +02002000 }
2001 }
Jens Axboe70524492006-04-11 15:51:17 +02002002
Jens Axboeaadd06e2006-07-10 11:00:01 +02002003 return ret;
Jens Axboe70524492006-04-11 15:51:17 +02002004}
2005
Heiko Carstens836f92a2009-01-14 14:14:33 +01002006SYSCALL_DEFINE4(tee, int, fdin, int, fdout, size_t, len, unsigned int, flags)
Jens Axboe70524492006-04-11 15:51:17 +02002007{
Al Viro2903ff02012-08-28 12:52:22 -04002008 struct fd in;
2009 int error;
Jens Axboe70524492006-04-11 15:51:17 +02002010
2011 if (unlikely(!len))
2012 return 0;
2013
2014 error = -EBADF;
Al Viro2903ff02012-08-28 12:52:22 -04002015 in = fdget(fdin);
2016 if (in.file) {
2017 if (in.file->f_mode & FMODE_READ) {
2018 struct fd out = fdget(fdout);
2019 if (out.file) {
2020 if (out.file->f_mode & FMODE_WRITE)
2021 error = do_tee(in.file, out.file,
2022 len, flags);
2023 fdput(out);
Jens Axboe70524492006-04-11 15:51:17 +02002024 }
2025 }
Al Viro2903ff02012-08-28 12:52:22 -04002026 fdput(in);
Jens Axboe70524492006-04-11 15:51:17 +02002027 }
2028
2029 return error;
2030}