blob: 29e394e49ddda7c7721d3939d993632a29f4499a [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 Viro06ae43f2013-03-20 13:19:30 -040034#include "internal.h"
Jens Axboe5274f052006-03-30 15:15:30 +020035
Jens Axboe83f91352006-04-02 23:05:09 +020036/*
37 * Attempt to steal a page from a pipe buffer. This should perhaps go into
38 * a vm helper function, it's already simplified quite a bit by the
39 * addition of remove_mapping(). If success is returned, the caller may
40 * attempt to reuse this page for another destination.
41 */
Jens Axboe76ad4d12006-05-03 10:41:33 +020042static int page_cache_pipe_buf_steal(struct pipe_inode_info *pipe,
Jens Axboe5abc97a2006-03-30 15:16:46 +020043 struct pipe_buffer *buf)
44{
45 struct page *page = buf->page;
Jens Axboe9e94cd42006-06-20 15:01:12 +020046 struct address_space *mapping;
Jens Axboe5abc97a2006-03-30 15:16:46 +020047
Jens Axboe9e0267c2006-04-19 15:57:31 +020048 lock_page(page);
49
Jens Axboe9e94cd42006-06-20 15:01:12 +020050 mapping = page_mapping(page);
51 if (mapping) {
52 WARN_ON(!PageUptodate(page));
Jens Axboe5abc97a2006-03-30 15:16:46 +020053
Jens Axboe9e94cd42006-06-20 15:01:12 +020054 /*
55 * At least for ext2 with nobh option, we need to wait on
56 * writeback completing on this page, since we'll remove it
57 * from the pagecache. Otherwise truncate wont wait on the
58 * page, allowing the disk blocks to be reused by someone else
59 * before we actually wrote our data to them. fs corruption
60 * ensues.
61 */
62 wait_on_page_writeback(page);
Jens Axboead8d6f02006-04-02 23:10:32 +020063
David Howells266cf652009-04-03 16:42:36 +010064 if (page_has_private(page) &&
65 !try_to_release_page(page, GFP_KERNEL))
Jens Axboeca39d652008-05-20 21:27:41 +020066 goto out_unlock;
Jens Axboe4f6f0bd2006-04-02 23:04:46 +020067
Jens Axboe9e94cd42006-06-20 15:01:12 +020068 /*
69 * If we succeeded in removing the mapping, set LRU flag
70 * and return good.
71 */
72 if (remove_mapping(mapping, page)) {
73 buf->flags |= PIPE_BUF_FLAG_LRU;
74 return 0;
75 }
Jens Axboe9e0267c2006-04-19 15:57:31 +020076 }
Jens Axboe5abc97a2006-03-30 15:16:46 +020077
Jens Axboe9e94cd42006-06-20 15:01:12 +020078 /*
79 * Raced with truncate or failed to remove page from current
80 * address space, unlock and return failure.
81 */
Jens Axboeca39d652008-05-20 21:27:41 +020082out_unlock:
Jens Axboe9e94cd42006-06-20 15:01:12 +020083 unlock_page(page);
84 return 1;
Jens Axboe5abc97a2006-03-30 15:16:46 +020085}
86
Jens Axboe76ad4d12006-05-03 10:41:33 +020087static void page_cache_pipe_buf_release(struct pipe_inode_info *pipe,
Jens Axboe5274f052006-03-30 15:15:30 +020088 struct pipe_buffer *buf)
89{
90 page_cache_release(buf->page);
Jens Axboe14328732006-05-03 10:35:26 +020091 buf->flags &= ~PIPE_BUF_FLAG_LRU;
Jens Axboe5274f052006-03-30 15:15:30 +020092}
93
Jens Axboe08457182007-06-12 20:51:32 +020094/*
95 * Check whether the contents of buf is OK to access. Since the content
96 * is a page cache page, IO may be in flight.
97 */
Jens Axboecac36bb02007-06-14 13:10:48 +020098static int page_cache_pipe_buf_confirm(struct pipe_inode_info *pipe,
99 struct pipe_buffer *buf)
Jens Axboe5274f052006-03-30 15:15:30 +0200100{
101 struct page *page = buf->page;
Jens Axboe49d0b212006-04-10 09:04:41 +0200102 int err;
Jens Axboe5274f052006-03-30 15:15:30 +0200103
104 if (!PageUptodate(page)) {
Jens Axboe49d0b212006-04-10 09:04:41 +0200105 lock_page(page);
106
107 /*
108 * Page got truncated/unhashed. This will cause a 0-byte
Ingo Molnar73d62d82006-04-11 13:57:21 +0200109 * splice, if this is the first page.
Jens Axboe49d0b212006-04-10 09:04:41 +0200110 */
111 if (!page->mapping) {
112 err = -ENODATA;
113 goto error;
114 }
115
116 /*
Ingo Molnar73d62d82006-04-11 13:57:21 +0200117 * Uh oh, read-error from disk.
Jens Axboe49d0b212006-04-10 09:04:41 +0200118 */
119 if (!PageUptodate(page)) {
120 err = -EIO;
121 goto error;
122 }
123
124 /*
Jens Axboef84d7512006-05-01 19:59:03 +0200125 * Page is ok afterall, we are done.
Jens Axboe49d0b212006-04-10 09:04:41 +0200126 */
Jens Axboe5274f052006-03-30 15:15:30 +0200127 unlock_page(page);
Jens Axboe5274f052006-03-30 15:15:30 +0200128 }
129
Jens Axboef84d7512006-05-01 19:59:03 +0200130 return 0;
Jens Axboe49d0b212006-04-10 09:04:41 +0200131error:
132 unlock_page(page);
Jens Axboef84d7512006-05-01 19:59:03 +0200133 return err;
Jens Axboe70524492006-04-11 15:51:17 +0200134}
135
Hugh Dickins708e3502011-07-25 17:12:32 -0700136const struct pipe_buf_operations page_cache_pipe_buf_ops = {
Jens Axboe5274f052006-03-30 15:15:30 +0200137 .can_merge = 0,
Jens Axboef84d7512006-05-01 19:59:03 +0200138 .map = generic_pipe_buf_map,
139 .unmap = generic_pipe_buf_unmap,
Jens Axboecac36bb02007-06-14 13:10:48 +0200140 .confirm = page_cache_pipe_buf_confirm,
Jens Axboe5274f052006-03-30 15:15:30 +0200141 .release = page_cache_pipe_buf_release,
Jens Axboe5abc97a2006-03-30 15:16:46 +0200142 .steal = page_cache_pipe_buf_steal,
Jens Axboef84d7512006-05-01 19:59:03 +0200143 .get = generic_pipe_buf_get,
Jens Axboe5274f052006-03-30 15:15:30 +0200144};
145
Jens Axboe912d35f2006-04-26 10:59:21 +0200146static int user_page_pipe_buf_steal(struct pipe_inode_info *pipe,
147 struct pipe_buffer *buf)
148{
Jens Axboe7afa6fd2006-05-01 20:02:33 +0200149 if (!(buf->flags & PIPE_BUF_FLAG_GIFT))
150 return 1;
151
Jens Axboe14328732006-05-03 10:35:26 +0200152 buf->flags |= PIPE_BUF_FLAG_LRU;
Jens Axboe330ab712006-05-02 15:29:57 +0200153 return generic_pipe_buf_steal(pipe, buf);
Jens Axboe912d35f2006-04-26 10:59:21 +0200154}
155
Eric Dumazetd4c3cca2006-12-13 00:34:04 -0800156static const struct pipe_buf_operations user_page_pipe_buf_ops = {
Jens Axboe912d35f2006-04-26 10:59:21 +0200157 .can_merge = 0,
Jens Axboef84d7512006-05-01 19:59:03 +0200158 .map = generic_pipe_buf_map,
159 .unmap = generic_pipe_buf_unmap,
Jens Axboecac36bb02007-06-14 13:10:48 +0200160 .confirm = generic_pipe_buf_confirm,
Jens Axboe912d35f2006-04-26 10:59:21 +0200161 .release = page_cache_pipe_buf_release,
162 .steal = user_page_pipe_buf_steal,
Jens Axboef84d7512006-05-01 19:59:03 +0200163 .get = generic_pipe_buf_get,
Jens Axboe912d35f2006-04-26 10:59:21 +0200164};
165
Namhyung Kim825cdcb2011-05-23 19:58:53 +0200166static void wakeup_pipe_readers(struct pipe_inode_info *pipe)
167{
168 smp_mb();
169 if (waitqueue_active(&pipe->wait))
170 wake_up_interruptible(&pipe->wait);
171 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
172}
173
Jens Axboe932cc6d2007-06-21 13:10:21 +0200174/**
175 * splice_to_pipe - fill passed data into a pipe
176 * @pipe: pipe to fill
177 * @spd: data to fill
178 *
179 * Description:
Randy Dunlap79685b82007-07-27 08:08:51 +0200180 * @spd contains a map of pages and len/offset tuples, along with
Jens Axboe932cc6d2007-06-21 13:10:21 +0200181 * the struct pipe_buf_operations associated with these pages. This
182 * function will link that data to the pipe.
183 *
Jens Axboe83f91352006-04-02 23:05:09 +0200184 */
Jens Axboed6b29d72007-06-04 09:59:47 +0200185ssize_t splice_to_pipe(struct pipe_inode_info *pipe,
186 struct splice_pipe_desc *spd)
Jens Axboe5274f052006-03-30 15:15:30 +0200187{
Jens Axboe00de00b2007-06-15 13:14:22 +0200188 unsigned int spd_pages = spd->nr_pages;
Jens Axboe912d35f2006-04-26 10:59:21 +0200189 int ret, do_wakeup, page_nr;
Jens Axboe5274f052006-03-30 15:15:30 +0200190
191 ret = 0;
192 do_wakeup = 0;
Jens Axboe912d35f2006-04-26 10:59:21 +0200193 page_nr = 0;
Jens Axboe5274f052006-03-30 15:15:30 +0200194
Miklos Szeredi61e0d472009-04-14 19:48:41 +0200195 pipe_lock(pipe);
Jens Axboe5274f052006-03-30 15:15:30 +0200196
Jens Axboe5274f052006-03-30 15:15:30 +0200197 for (;;) {
Ingo Molnar3a326a22006-04-10 15:18:35 +0200198 if (!pipe->readers) {
Jens Axboe5274f052006-03-30 15:15:30 +0200199 send_sig(SIGPIPE, current, 0);
200 if (!ret)
201 ret = -EPIPE;
202 break;
203 }
204
Jens Axboe35f3d142010-05-20 10:43:18 +0200205 if (pipe->nrbufs < pipe->buffers) {
206 int newbuf = (pipe->curbuf + pipe->nrbufs) & (pipe->buffers - 1);
Ingo Molnar3a326a22006-04-10 15:18:35 +0200207 struct pipe_buffer *buf = pipe->bufs + newbuf;
Jens Axboe5274f052006-03-30 15:15:30 +0200208
Jens Axboe912d35f2006-04-26 10:59:21 +0200209 buf->page = spd->pages[page_nr];
210 buf->offset = spd->partial[page_nr].offset;
211 buf->len = spd->partial[page_nr].len;
Jens Axboe497f9622007-06-11 12:00:45 +0200212 buf->private = spd->partial[page_nr].private;
Jens Axboe912d35f2006-04-26 10:59:21 +0200213 buf->ops = spd->ops;
Jens Axboe7afa6fd2006-05-01 20:02:33 +0200214 if (spd->flags & SPLICE_F_GIFT)
215 buf->flags |= PIPE_BUF_FLAG_GIFT;
216
Jens Axboe6f767b02006-04-11 13:53:56 +0200217 pipe->nrbufs++;
Jens Axboe912d35f2006-04-26 10:59:21 +0200218 page_nr++;
219 ret += buf->len;
220
Jens Axboe6f767b02006-04-11 13:53:56 +0200221 if (pipe->inode)
222 do_wakeup = 1;
Jens Axboe5274f052006-03-30 15:15:30 +0200223
Jens Axboe912d35f2006-04-26 10:59:21 +0200224 if (!--spd->nr_pages)
Jens Axboe5274f052006-03-30 15:15:30 +0200225 break;
Jens Axboe35f3d142010-05-20 10:43:18 +0200226 if (pipe->nrbufs < pipe->buffers)
Jens Axboe5274f052006-03-30 15:15:30 +0200227 continue;
228
229 break;
230 }
231
Jens Axboe912d35f2006-04-26 10:59:21 +0200232 if (spd->flags & SPLICE_F_NONBLOCK) {
Linus Torvalds29e35092006-04-02 12:46:35 -0700233 if (!ret)
234 ret = -EAGAIN;
235 break;
236 }
237
Jens Axboe5274f052006-03-30 15:15:30 +0200238 if (signal_pending(current)) {
239 if (!ret)
240 ret = -ERESTARTSYS;
241 break;
242 }
243
244 if (do_wakeup) {
Jens Axboec0bd1f62006-04-10 09:03:32 +0200245 smp_mb();
Ingo Molnar3a326a22006-04-10 15:18:35 +0200246 if (waitqueue_active(&pipe->wait))
247 wake_up_interruptible_sync(&pipe->wait);
248 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
Jens Axboe5274f052006-03-30 15:15:30 +0200249 do_wakeup = 0;
250 }
251
Ingo Molnar3a326a22006-04-10 15:18:35 +0200252 pipe->waiting_writers++;
253 pipe_wait(pipe);
254 pipe->waiting_writers--;
Jens Axboe5274f052006-03-30 15:15:30 +0200255 }
256
Miklos Szeredi61e0d472009-04-14 19:48:41 +0200257 pipe_unlock(pipe);
Jens Axboe5274f052006-03-30 15:15:30 +0200258
Namhyung Kim825cdcb2011-05-23 19:58:53 +0200259 if (do_wakeup)
260 wakeup_pipe_readers(pipe);
Jens Axboe5274f052006-03-30 15:15:30 +0200261
Jens Axboe00de00b2007-06-15 13:14:22 +0200262 while (page_nr < spd_pages)
Jens Axboebbdfc2f2007-11-06 23:29:47 -0800263 spd->spd_release(spd, page_nr++);
Jens Axboe5274f052006-03-30 15:15:30 +0200264
265 return ret;
266}
267
Hugh Dickins708e3502011-07-25 17:12:32 -0700268void spd_release_page(struct splice_pipe_desc *spd, unsigned int i)
Jens Axboebbdfc2f2007-11-06 23:29:47 -0800269{
270 page_cache_release(spd->pages[i]);
271}
272
Jens Axboe35f3d142010-05-20 10:43:18 +0200273/*
274 * Check if we need to grow the arrays holding pages and partial page
275 * descriptions.
276 */
Eric Dumazet047fe362012-06-12 15:24:40 +0200277int splice_grow_spd(const struct pipe_inode_info *pipe, struct splice_pipe_desc *spd)
Jens Axboe35f3d142010-05-20 10:43:18 +0200278{
Eric Dumazet047fe362012-06-12 15:24:40 +0200279 unsigned int buffers = ACCESS_ONCE(pipe->buffers);
280
281 spd->nr_pages_max = buffers;
282 if (buffers <= PIPE_DEF_BUFFERS)
Jens Axboe35f3d142010-05-20 10:43:18 +0200283 return 0;
284
Eric Dumazet047fe362012-06-12 15:24:40 +0200285 spd->pages = kmalloc(buffers * sizeof(struct page *), GFP_KERNEL);
286 spd->partial = kmalloc(buffers * sizeof(struct partial_page), GFP_KERNEL);
Jens Axboe35f3d142010-05-20 10:43:18 +0200287
288 if (spd->pages && spd->partial)
289 return 0;
290
291 kfree(spd->pages);
292 kfree(spd->partial);
293 return -ENOMEM;
294}
295
Eric Dumazet047fe362012-06-12 15:24:40 +0200296void splice_shrink_spd(struct splice_pipe_desc *spd)
Jens Axboe35f3d142010-05-20 10:43:18 +0200297{
Eric Dumazet047fe362012-06-12 15:24:40 +0200298 if (spd->nr_pages_max <= PIPE_DEF_BUFFERS)
Jens Axboe35f3d142010-05-20 10:43:18 +0200299 return;
300
301 kfree(spd->pages);
302 kfree(spd->partial);
303}
304
Ingo Molnar3a326a22006-04-10 15:18:35 +0200305static int
Jens Axboecbb7e572006-04-11 14:57:50 +0200306__generic_file_splice_read(struct file *in, loff_t *ppos,
307 struct pipe_inode_info *pipe, size_t len,
308 unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +0200309{
310 struct address_space *mapping = in->f_mapping;
Fengguang Wud8983912007-07-19 01:48:06 -0700311 unsigned int loff, nr_pages, req_pages;
Jens Axboe35f3d142010-05-20 10:43:18 +0200312 struct page *pages[PIPE_DEF_BUFFERS];
313 struct partial_page partial[PIPE_DEF_BUFFERS];
Jens Axboe5274f052006-03-30 15:15:30 +0200314 struct page *page;
Jens Axboe91ad66e2006-04-19 15:55:10 +0200315 pgoff_t index, end_index;
316 loff_t isize;
Jens Axboeeb207962006-04-27 11:05:22 +0200317 int error, page_nr;
Jens Axboe912d35f2006-04-26 10:59:21 +0200318 struct splice_pipe_desc spd = {
319 .pages = pages,
320 .partial = partial,
Eric Dumazet047fe362012-06-12 15:24:40 +0200321 .nr_pages_max = PIPE_DEF_BUFFERS,
Jens Axboe912d35f2006-04-26 10:59:21 +0200322 .flags = flags,
323 .ops = &page_cache_pipe_buf_ops,
Jens Axboebbdfc2f2007-11-06 23:29:47 -0800324 .spd_release = spd_release_page,
Jens Axboe912d35f2006-04-26 10:59:21 +0200325 };
Jens Axboe5274f052006-03-30 15:15:30 +0200326
Jens Axboe35f3d142010-05-20 10:43:18 +0200327 if (splice_grow_spd(pipe, &spd))
328 return -ENOMEM;
329
Jens Axboecbb7e572006-04-11 14:57:50 +0200330 index = *ppos >> PAGE_CACHE_SHIFT;
Jens Axboe912d35f2006-04-26 10:59:21 +0200331 loff = *ppos & ~PAGE_CACHE_MASK;
Fengguang Wud8983912007-07-19 01:48:06 -0700332 req_pages = (len + loff + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
Eric Dumazet047fe362012-06-12 15:24:40 +0200333 nr_pages = min(req_pages, spd.nr_pages_max);
Jens Axboe5274f052006-03-30 15:15:30 +0200334
335 /*
Jens Axboeeb207962006-04-27 11:05:22 +0200336 * Lookup the (hopefully) full range of pages we need.
337 */
Jens Axboe35f3d142010-05-20 10:43:18 +0200338 spd.nr_pages = find_get_pages_contig(mapping, index, nr_pages, spd.pages);
Fengguang Wu431a48202007-07-19 01:48:05 -0700339 index += spd.nr_pages;
Jens Axboeeb207962006-04-27 11:05:22 +0200340
341 /*
342 * If find_get_pages_contig() returned fewer pages than we needed,
Fengguang Wu431a48202007-07-19 01:48:05 -0700343 * readahead/allocate the rest and fill in the holes.
Jens Axboeeb207962006-04-27 11:05:22 +0200344 */
Fengguang Wu431a48202007-07-19 01:48:05 -0700345 if (spd.nr_pages < nr_pages)
Rusty Russellcf914a72007-07-19 01:48:08 -0700346 page_cache_sync_readahead(mapping, &in->f_ra, in,
347 index, req_pages - spd.nr_pages);
Fengguang Wu431a48202007-07-19 01:48:05 -0700348
Jens Axboe932cc6d2007-06-21 13:10:21 +0200349 error = 0;
Jens Axboeeb207962006-04-27 11:05:22 +0200350 while (spd.nr_pages < nr_pages) {
351 /*
352 * Page could be there, find_get_pages_contig() breaks on
353 * the first hole.
354 */
355 page = find_get_page(mapping, index);
356 if (!page) {
Jens Axboee27dedd2006-05-01 19:59:54 +0200357 /*
Jens Axboeeb207962006-04-27 11:05:22 +0200358 * page didn't exist, allocate one.
359 */
360 page = page_cache_alloc_cold(mapping);
361 if (!page)
362 break;
363
364 error = add_to_page_cache_lru(page, mapping, index,
Nick Piggin0ae0b5d2010-05-25 10:25:26 +0200365 GFP_KERNEL);
Jens Axboeeb207962006-04-27 11:05:22 +0200366 if (unlikely(error)) {
367 page_cache_release(page);
Jens Axboea0548872006-05-03 10:58:22 +0200368 if (error == -EEXIST)
369 continue;
Jens Axboeeb207962006-04-27 11:05:22 +0200370 break;
371 }
372 /*
373 * add_to_page_cache() locks the page, unlock it
374 * to avoid convoluting the logic below even more.
375 */
376 unlock_page(page);
377 }
378
Jens Axboe35f3d142010-05-20 10:43:18 +0200379 spd.pages[spd.nr_pages++] = page;
Jens Axboeeb207962006-04-27 11:05:22 +0200380 index++;
381 }
382
383 /*
384 * Now loop over the map and see if we need to start IO on any
385 * pages, fill in the partial map, etc.
386 */
387 index = *ppos >> PAGE_CACHE_SHIFT;
388 nr_pages = spd.nr_pages;
389 spd.nr_pages = 0;
390 for (page_nr = 0; page_nr < nr_pages; page_nr++) {
Jens Axboe82aa5d62006-04-20 13:05:48 +0200391 unsigned int this_len;
392
393 if (!len)
394 break;
395
396 /*
397 * this_len is the max we'll use from this page
398 */
Andrew Mortonba5f5d92006-04-25 15:33:34 +0200399 this_len = min_t(unsigned long, len, PAGE_CACHE_SIZE - loff);
Jens Axboe35f3d142010-05-20 10:43:18 +0200400 page = spd.pages[page_nr];
Jens Axboe7480a902006-04-11 13:52:47 +0200401
Fengguang Wua08a1662007-07-19 01:48:03 -0700402 if (PageReadahead(page))
Rusty Russellcf914a72007-07-19 01:48:08 -0700403 page_cache_async_readahead(mapping, &in->f_ra, in,
Fengguang Wud8983912007-07-19 01:48:06 -0700404 page, index, req_pages - page_nr);
Fengguang Wua08a1662007-07-19 01:48:03 -0700405
Jens Axboe7480a902006-04-11 13:52:47 +0200406 /*
407 * If the page isn't uptodate, we may need to start io on it
408 */
409 if (!PageUptodate(page)) {
Miklos Szeredi69650312010-08-03 12:48:50 +0200410 lock_page(page);
Jens Axboe7480a902006-04-11 13:52:47 +0200411
412 /*
Miklos Szeredi32502b82008-07-04 09:35:17 +0200413 * Page was truncated, or invalidated by the
414 * filesystem. Redo the find/create, but this time the
415 * page is kept locked, so there's no chance of another
416 * race with truncate/invalidate.
Jens Axboe7480a902006-04-11 13:52:47 +0200417 */
418 if (!page->mapping) {
419 unlock_page(page);
Miklos Szeredi32502b82008-07-04 09:35:17 +0200420 page = find_or_create_page(mapping, index,
421 mapping_gfp_mask(mapping));
422
423 if (!page) {
424 error = -ENOMEM;
425 break;
426 }
Jens Axboe35f3d142010-05-20 10:43:18 +0200427 page_cache_release(spd.pages[page_nr]);
428 spd.pages[page_nr] = page;
Jens Axboe7480a902006-04-11 13:52:47 +0200429 }
430 /*
431 * page was already under io and is now done, great
432 */
433 if (PageUptodate(page)) {
434 unlock_page(page);
435 goto fill_it;
436 }
437
Jens Axboe7480a902006-04-11 13:52:47 +0200438 /*
439 * need to read in the page
440 */
441 error = mapping->a_ops->readpage(in, page);
Jens Axboe7480a902006-04-11 13:52:47 +0200442 if (unlikely(error)) {
Jens Axboeeb207962006-04-27 11:05:22 +0200443 /*
444 * We really should re-lookup the page here,
445 * but it complicates things a lot. Instead
446 * lets just do what we already stored, and
447 * we'll get it the next time we are called.
448 */
Jens Axboe7480a902006-04-11 13:52:47 +0200449 if (error == AOP_TRUNCATED_PAGE)
Jens Axboeeb207962006-04-27 11:05:22 +0200450 error = 0;
451
Jens Axboe7480a902006-04-11 13:52:47 +0200452 break;
453 }
Jens Axboe620a3242007-06-07 09:39:42 +0200454 }
455fill_it:
456 /*
457 * i_size must be checked after PageUptodate.
458 */
459 isize = i_size_read(mapping->host);
460 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
461 if (unlikely(!isize || index > end_index))
462 break;
463
464 /*
465 * if this is the last page, see if we need to shrink
466 * the length and stop
467 */
468 if (end_index == index) {
469 unsigned int plen;
Jens Axboe91ad66e2006-04-19 15:55:10 +0200470
471 /*
Jens Axboe620a3242007-06-07 09:39:42 +0200472 * max good bytes in this page
Jens Axboe91ad66e2006-04-19 15:55:10 +0200473 */
Jens Axboe620a3242007-06-07 09:39:42 +0200474 plen = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
475 if (plen <= loff)
Jens Axboe91ad66e2006-04-19 15:55:10 +0200476 break;
Jens Axboe91ad66e2006-04-19 15:55:10 +0200477
478 /*
Jens Axboe620a3242007-06-07 09:39:42 +0200479 * force quit after adding this page
Jens Axboe91ad66e2006-04-19 15:55:10 +0200480 */
Jens Axboe620a3242007-06-07 09:39:42 +0200481 this_len = min(this_len, plen - loff);
482 len = this_len;
Jens Axboe7480a902006-04-11 13:52:47 +0200483 }
Jens Axboe620a3242007-06-07 09:39:42 +0200484
Jens Axboe35f3d142010-05-20 10:43:18 +0200485 spd.partial[page_nr].offset = loff;
486 spd.partial[page_nr].len = this_len;
Jens Axboe82aa5d62006-04-20 13:05:48 +0200487 len -= this_len;
Jens Axboe91ad66e2006-04-19 15:55:10 +0200488 loff = 0;
Jens Axboeeb207962006-04-27 11:05:22 +0200489 spd.nr_pages++;
490 index++;
Jens Axboe5274f052006-03-30 15:15:30 +0200491 }
492
Jens Axboeeb207962006-04-27 11:05:22 +0200493 /*
Hugh Dickins475ecad2007-06-07 09:36:00 +0200494 * Release any pages at the end, if we quit early. 'page_nr' is how far
Jens Axboeeb207962006-04-27 11:05:22 +0200495 * we got, 'nr_pages' is how many pages are in the map.
496 */
497 while (page_nr < nr_pages)
Jens Axboe35f3d142010-05-20 10:43:18 +0200498 page_cache_release(spd.pages[page_nr++]);
Fengguang Wuf4e6b492007-10-16 01:24:33 -0700499 in->f_ra.prev_pos = (loff_t)index << PAGE_CACHE_SHIFT;
Jens Axboeeb207962006-04-27 11:05:22 +0200500
Jens Axboe912d35f2006-04-26 10:59:21 +0200501 if (spd.nr_pages)
Jens Axboe35f3d142010-05-20 10:43:18 +0200502 error = splice_to_pipe(pipe, &spd);
Jens Axboe5274f052006-03-30 15:15:30 +0200503
Eric Dumazet047fe362012-06-12 15:24:40 +0200504 splice_shrink_spd(&spd);
Jens Axboe7480a902006-04-11 13:52:47 +0200505 return error;
Jens Axboe5274f052006-03-30 15:15:30 +0200506}
507
Jens Axboe83f91352006-04-02 23:05:09 +0200508/**
509 * generic_file_splice_read - splice data from file to a pipe
510 * @in: file to splice from
Jens Axboe932cc6d2007-06-21 13:10:21 +0200511 * @ppos: position in @in
Jens Axboe83f91352006-04-02 23:05:09 +0200512 * @pipe: pipe to splice to
513 * @len: number of bytes to splice
514 * @flags: splice modifier flags
515 *
Jens Axboe932cc6d2007-06-21 13:10:21 +0200516 * Description:
517 * Will read pages from given file and fill them into a pipe. Can be
518 * used as long as the address_space operations for the source implements
519 * a readpage() hook.
520 *
Jens Axboe83f91352006-04-02 23:05:09 +0200521 */
Jens Axboecbb7e572006-04-11 14:57:50 +0200522ssize_t generic_file_splice_read(struct file *in, loff_t *ppos,
523 struct pipe_inode_info *pipe, size_t len,
524 unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +0200525{
Jens Axboed366d3982007-06-01 14:54:11 +0200526 loff_t isize, left;
Jens Axboe8191ecd2008-04-10 08:24:25 +0200527 int ret;
Jens Axboed366d3982007-06-01 14:54:11 +0200528
529 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,
549 .map = generic_pipe_buf_map,
550 .unmap = generic_pipe_buf_unmap,
551 .confirm = generic_pipe_buf_confirm,
552 .release = generic_pipe_buf_release,
553 .steal = generic_pipe_buf_steal,
554 .get = generic_pipe_buf_get,
555};
556
557static ssize_t kernel_readv(struct file *file, const struct iovec *vec,
558 unsigned long vlen, loff_t offset)
559{
560 mm_segment_t old_fs;
561 loff_t pos = offset;
562 ssize_t res;
563
564 old_fs = get_fs();
565 set_fs(get_ds());
566 /* The cast to a user pointer is valid due to the set_fs() */
567 res = vfs_readv(file, (const struct iovec __user *)vec, vlen, &pos);
568 set_fs(old_fs);
569
570 return res;
571}
572
Al Viro7bb307e2013-02-23 14:51:48 -0500573ssize_t kernel_write(struct file *file, const char *buf, size_t count,
Miklos Szeredib2858d72009-05-19 11:37:46 +0200574 loff_t pos)
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +0200575{
576 mm_segment_t old_fs;
577 ssize_t res;
578
579 old_fs = get_fs();
580 set_fs(get_ds());
581 /* The cast to a user pointer is valid due to the set_fs() */
Al Viro7bb307e2013-02-23 14:51:48 -0500582 res = vfs_write(file, (__force const char __user *)buf, count, &pos);
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +0200583 set_fs(old_fs);
584
585 return res;
586}
Al Viro7bb307e2013-02-23 14:51:48 -0500587EXPORT_SYMBOL(kernel_write);
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +0200588
Miklos Szeredi68181732009-05-07 15:37:36 +0200589ssize_t default_file_splice_read(struct file *in, loff_t *ppos,
590 struct pipe_inode_info *pipe, size_t len,
591 unsigned int flags)
592{
593 unsigned int nr_pages;
594 unsigned int nr_freed;
595 size_t offset;
Jens Axboe35f3d142010-05-20 10:43:18 +0200596 struct page *pages[PIPE_DEF_BUFFERS];
597 struct partial_page partial[PIPE_DEF_BUFFERS];
598 struct iovec *vec, __vec[PIPE_DEF_BUFFERS];
Miklos Szeredi68181732009-05-07 15:37:36 +0200599 ssize_t res;
600 size_t this_len;
601 int error;
602 int i;
603 struct splice_pipe_desc spd = {
604 .pages = pages,
605 .partial = partial,
Eric Dumazet047fe362012-06-12 15:24:40 +0200606 .nr_pages_max = PIPE_DEF_BUFFERS,
Miklos Szeredi68181732009-05-07 15:37:36 +0200607 .flags = flags,
608 .ops = &default_pipe_buf_ops,
609 .spd_release = spd_release_page,
610 };
611
Jens Axboe35f3d142010-05-20 10:43:18 +0200612 if (splice_grow_spd(pipe, &spd))
613 return -ENOMEM;
614
615 res = -ENOMEM;
616 vec = __vec;
Eric Dumazet047fe362012-06-12 15:24:40 +0200617 if (spd.nr_pages_max > PIPE_DEF_BUFFERS) {
618 vec = kmalloc(spd.nr_pages_max * sizeof(struct iovec), GFP_KERNEL);
Jens Axboe35f3d142010-05-20 10:43:18 +0200619 if (!vec)
620 goto shrink_ret;
621 }
622
Miklos Szeredi68181732009-05-07 15:37:36 +0200623 offset = *ppos & ~PAGE_CACHE_MASK;
624 nr_pages = (len + offset + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
625
Eric Dumazet047fe362012-06-12 15:24:40 +0200626 for (i = 0; i < nr_pages && i < spd.nr_pages_max && len; i++) {
Miklos Szeredi68181732009-05-07 15:37:36 +0200627 struct page *page;
628
Jens Axboe4f231222009-05-13 08:35:35 +0200629 page = alloc_page(GFP_USER);
Miklos Szeredi68181732009-05-07 15:37:36 +0200630 error = -ENOMEM;
631 if (!page)
632 goto err;
633
634 this_len = min_t(size_t, len, PAGE_CACHE_SIZE - offset);
Jens Axboe4f231222009-05-13 08:35:35 +0200635 vec[i].iov_base = (void __user *) page_address(page);
Miklos Szeredi68181732009-05-07 15:37:36 +0200636 vec[i].iov_len = this_len;
Jens Axboe35f3d142010-05-20 10:43:18 +0200637 spd.pages[i] = page;
Miklos Szeredi68181732009-05-07 15:37:36 +0200638 spd.nr_pages++;
639 len -= this_len;
640 offset = 0;
641 }
642
643 res = kernel_readv(in, vec, spd.nr_pages, *ppos);
Andrew Morton77f6bf52009-05-14 09:49:44 +0200644 if (res < 0) {
645 error = res;
Miklos Szeredi68181732009-05-07 15:37:36 +0200646 goto err;
Andrew Morton77f6bf52009-05-14 09:49:44 +0200647 }
Miklos Szeredi68181732009-05-07 15:37:36 +0200648
649 error = 0;
650 if (!res)
651 goto err;
652
653 nr_freed = 0;
654 for (i = 0; i < spd.nr_pages; i++) {
Miklos Szeredi68181732009-05-07 15:37:36 +0200655 this_len = min_t(size_t, vec[i].iov_len, res);
Jens Axboe35f3d142010-05-20 10:43:18 +0200656 spd.partial[i].offset = 0;
657 spd.partial[i].len = this_len;
Miklos Szeredi68181732009-05-07 15:37:36 +0200658 if (!this_len) {
Jens Axboe35f3d142010-05-20 10:43:18 +0200659 __free_page(spd.pages[i]);
660 spd.pages[i] = NULL;
Miklos Szeredi68181732009-05-07 15:37:36 +0200661 nr_freed++;
662 }
663 res -= this_len;
664 }
665 spd.nr_pages -= nr_freed;
666
667 res = splice_to_pipe(pipe, &spd);
668 if (res > 0)
669 *ppos += res;
670
Jens Axboe35f3d142010-05-20 10:43:18 +0200671shrink_ret:
672 if (vec != __vec)
673 kfree(vec);
Eric Dumazet047fe362012-06-12 15:24:40 +0200674 splice_shrink_spd(&spd);
Miklos Szeredi68181732009-05-07 15:37:36 +0200675 return res;
676
677err:
Jens Axboe4f231222009-05-13 08:35:35 +0200678 for (i = 0; i < spd.nr_pages; i++)
Jens Axboe35f3d142010-05-20 10:43:18 +0200679 __free_page(spd.pages[i]);
Jens Axboe4f231222009-05-13 08:35:35 +0200680
Jens Axboe35f3d142010-05-20 10:43:18 +0200681 res = error;
682 goto shrink_ret;
Miklos Szeredi68181732009-05-07 15:37:36 +0200683}
684EXPORT_SYMBOL(default_file_splice_read);
685
Jens Axboe5274f052006-03-30 15:15:30 +0200686/*
Jens Axboe4f6f0bd2006-04-02 23:04:46 +0200687 * Send 'sd->len' bytes to socket from 'sd->file' at position 'sd->pos'
Jens Axboe016b6612006-04-25 15:42:00 +0200688 * using sendpage(). Return the number of bytes sent.
Jens Axboe5274f052006-03-30 15:15:30 +0200689 */
Jens Axboe76ad4d12006-05-03 10:41:33 +0200690static int pipe_to_sendpage(struct pipe_inode_info *pipe,
Jens Axboe5274f052006-03-30 15:15:30 +0200691 struct pipe_buffer *buf, struct splice_desc *sd)
692{
Jens Axboe6a14b902007-06-14 13:08:55 +0200693 struct file *file = sd->u.file;
Jens Axboe5274f052006-03-30 15:15:30 +0200694 loff_t pos = sd->pos;
Michał Mirosława8adbe32010-12-17 08:56:44 +0100695 int more;
Jens Axboe5274f052006-03-30 15:15:30 +0200696
Michał Mirosława8adbe32010-12-17 08:56:44 +0100697 if (!likely(file->f_op && file->f_op->sendpage))
698 return -EINVAL;
Jens Axboe5274f052006-03-30 15:15:30 +0200699
Eric Dumazet35f9c092012-04-05 03:05:35 +0000700 more = (sd->flags & SPLICE_F_MORE) ? MSG_MORE : 0;
Eric Dumazetae62ca72013-01-06 18:21:49 +0000701
702 if (sd->len < sd->total_len && pipe->nrbufs > 1)
Eric Dumazet35f9c092012-04-05 03:05:35 +0000703 more |= MSG_SENDPAGE_NOTLAST;
Eric Dumazetae62ca72013-01-06 18:21:49 +0000704
Michał Mirosława8adbe32010-12-17 08:56:44 +0100705 return file->f_op->sendpage(file, buf->page, buf->offset,
706 sd->len, &pos, more);
Jens Axboe5274f052006-03-30 15:15:30 +0200707}
708
709/*
710 * This is a little more tricky than the file -> pipe splicing. There are
711 * basically three cases:
712 *
713 * - Destination page already exists in the address space and there
714 * are users of it. For that case we have no other option that
715 * copying the data. Tough luck.
716 * - Destination page already exists in the address space, but there
717 * are no users of it. Make sure it's uptodate, then drop it. Fall
718 * through to last case.
719 * - Destination page does not exist, we can add the pipe page to
720 * the page cache and avoid the copy.
721 *
Jens Axboe83f91352006-04-02 23:05:09 +0200722 * If asked to move pages to the output file (SPLICE_F_MOVE is set in
723 * sd->flags), we attempt to migrate pages from the pipe to the output
724 * file address space page cache. This is possible if no one else has
725 * the pipe page referenced outside of the pipe and page cache. If
726 * SPLICE_F_MOVE isn't set, or we cannot move the page, we simply create
727 * a new page in the output file page cache and fill/dirty that.
Jens Axboe5274f052006-03-30 15:15:30 +0200728 */
Miklos Szeredi328eaab2009-04-14 19:48:39 +0200729int pipe_to_file(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
730 struct splice_desc *sd)
Jens Axboe5274f052006-03-30 15:15:30 +0200731{
Jens Axboe6a14b902007-06-14 13:08:55 +0200732 struct file *file = sd->u.file;
Jens Axboe5274f052006-03-30 15:15:30 +0200733 struct address_space *mapping = file->f_mapping;
Jens Axboe016b6612006-04-25 15:42:00 +0200734 unsigned int offset, this_len;
Jens Axboe5274f052006-03-30 15:15:30 +0200735 struct page *page;
Nick Pigginafddba42007-10-16 01:25:01 -0700736 void *fsdata;
Jens Axboe3e7ee3e2006-04-02 23:11:04 +0200737 int ret;
Jens Axboe5274f052006-03-30 15:15:30 +0200738
Jens Axboe5274f052006-03-30 15:15:30 +0200739 offset = sd->pos & ~PAGE_CACHE_MASK;
740
Jens Axboe016b6612006-04-25 15:42:00 +0200741 this_len = sd->len;
742 if (this_len + offset > PAGE_CACHE_SIZE)
743 this_len = PAGE_CACHE_SIZE - offset;
744
Nick Pigginafddba42007-10-16 01:25:01 -0700745 ret = pagecache_write_begin(file, mapping, sd->pos, this_len,
746 AOP_FLAG_UNINTERRUPTIBLE, &page, &fsdata);
747 if (unlikely(ret))
748 goto out;
Jens Axboe5274f052006-03-30 15:15:30 +0200749
Jens Axboe0568b402006-05-01 19:50:48 +0200750 if (buf->page != page) {
Jens Axboe76ad4d12006-05-03 10:41:33 +0200751 char *src = buf->ops->map(pipe, buf, 1);
Cong Wange8e3c3d2011-11-25 23:14:27 +0800752 char *dst = kmap_atomic(page);
Jens Axboe5abc97a2006-03-30 15:16:46 +0200753
Jens Axboe016b6612006-04-25 15:42:00 +0200754 memcpy(dst + offset, src + buf->offset, this_len);
Jens Axboe5abc97a2006-03-30 15:16:46 +0200755 flush_dcache_page(page);
Cong Wange8e3c3d2011-11-25 23:14:27 +0800756 kunmap_atomic(dst);
Jens Axboe76ad4d12006-05-03 10:41:33 +0200757 buf->ops->unmap(pipe, buf, src);
Jens Axboe5abc97a2006-03-30 15:16:46 +0200758 }
Nick Pigginafddba42007-10-16 01:25:01 -0700759 ret = pagecache_write_end(file, mapping, sd->pos, this_len, this_len,
760 page, fsdata);
Jens Axboe5274f052006-03-30 15:15:30 +0200761out:
Jens Axboe5274f052006-03-30 15:15:30 +0200762 return ret;
763}
Miklos Szeredi328eaab2009-04-14 19:48:39 +0200764EXPORT_SYMBOL(pipe_to_file);
Jens Axboe5274f052006-03-30 15:15:30 +0200765
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200766static void wakeup_pipe_writers(struct pipe_inode_info *pipe)
767{
768 smp_mb();
769 if (waitqueue_active(&pipe->wait))
770 wake_up_interruptible(&pipe->wait);
771 kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
772}
773
774/**
775 * splice_from_pipe_feed - feed available data from a pipe to a file
776 * @pipe: pipe to splice from
777 * @sd: information to @actor
778 * @actor: handler that splices the data
779 *
780 * Description:
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200781 * This function loops over the pipe and calls @actor to do the
782 * actual moving of a single struct pipe_buffer to the desired
783 * destination. It returns when there's no more buffers left in
784 * the pipe or if the requested number of bytes (@sd->total_len)
785 * have been copied. It returns a positive number (one) if the
786 * pipe needs to be filled with more data, zero if the required
787 * number of bytes have been copied and -errno on error.
788 *
789 * This, together with splice_from_pipe_{begin,end,next}, may be
790 * used to implement the functionality of __splice_from_pipe() when
791 * locking is required around copying the pipe buffers to the
792 * destination.
793 */
794int splice_from_pipe_feed(struct pipe_inode_info *pipe, struct splice_desc *sd,
795 splice_actor *actor)
796{
797 int ret;
798
799 while (pipe->nrbufs) {
800 struct pipe_buffer *buf = pipe->bufs + pipe->curbuf;
801 const struct pipe_buf_operations *ops = buf->ops;
802
803 sd->len = buf->len;
804 if (sd->len > sd->total_len)
805 sd->len = sd->total_len;
806
Michał Mirosława8adbe32010-12-17 08:56:44 +0100807 ret = buf->ops->confirm(pipe, buf);
808 if (unlikely(ret)) {
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200809 if (ret == -ENODATA)
810 ret = 0;
811 return ret;
812 }
Michał Mirosława8adbe32010-12-17 08:56:44 +0100813
814 ret = actor(pipe, buf, sd);
815 if (ret <= 0)
816 return ret;
817
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200818 buf->offset += ret;
819 buf->len -= ret;
820
821 sd->num_spliced += ret;
822 sd->len -= ret;
823 sd->pos += ret;
824 sd->total_len -= ret;
825
826 if (!buf->len) {
827 buf->ops = NULL;
828 ops->release(pipe, buf);
Jens Axboe35f3d142010-05-20 10:43:18 +0200829 pipe->curbuf = (pipe->curbuf + 1) & (pipe->buffers - 1);
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200830 pipe->nrbufs--;
831 if (pipe->inode)
832 sd->need_wakeup = true;
833 }
834
835 if (!sd->total_len)
836 return 0;
837 }
838
839 return 1;
840}
841EXPORT_SYMBOL(splice_from_pipe_feed);
842
843/**
844 * splice_from_pipe_next - wait for some data to splice from
845 * @pipe: pipe to splice from
846 * @sd: information about the splice operation
847 *
848 * Description:
849 * This function will wait for some data and return a positive
850 * value (one) if pipe buffers are available. It will return zero
851 * or -errno if no more data needs to be spliced.
852 */
853int splice_from_pipe_next(struct pipe_inode_info *pipe, struct splice_desc *sd)
854{
855 while (!pipe->nrbufs) {
856 if (!pipe->writers)
857 return 0;
858
859 if (!pipe->waiting_writers && sd->num_spliced)
860 return 0;
861
862 if (sd->flags & SPLICE_F_NONBLOCK)
863 return -EAGAIN;
864
865 if (signal_pending(current))
866 return -ERESTARTSYS;
867
868 if (sd->need_wakeup) {
869 wakeup_pipe_writers(pipe);
870 sd->need_wakeup = false;
871 }
872
873 pipe_wait(pipe);
874 }
875
876 return 1;
877}
878EXPORT_SYMBOL(splice_from_pipe_next);
879
880/**
881 * splice_from_pipe_begin - start splicing from pipe
Randy Dunlapb80901b2009-04-16 19:09:55 -0700882 * @sd: information about the splice operation
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200883 *
884 * Description:
885 * This function should be called before a loop containing
886 * splice_from_pipe_next() and splice_from_pipe_feed() to
887 * initialize the necessary fields of @sd.
888 */
889void splice_from_pipe_begin(struct splice_desc *sd)
890{
891 sd->num_spliced = 0;
892 sd->need_wakeup = false;
893}
894EXPORT_SYMBOL(splice_from_pipe_begin);
895
896/**
897 * splice_from_pipe_end - finish splicing from pipe
898 * @pipe: pipe to splice from
899 * @sd: information about the splice operation
900 *
901 * Description:
902 * This function will wake up pipe writers if necessary. It should
903 * be called after a loop containing splice_from_pipe_next() and
904 * splice_from_pipe_feed().
905 */
906void splice_from_pipe_end(struct pipe_inode_info *pipe, struct splice_desc *sd)
907{
908 if (sd->need_wakeup)
909 wakeup_pipe_writers(pipe);
910}
911EXPORT_SYMBOL(splice_from_pipe_end);
912
Jens Axboe932cc6d2007-06-21 13:10:21 +0200913/**
914 * __splice_from_pipe - splice data from a pipe to given actor
915 * @pipe: pipe to splice from
916 * @sd: information to @actor
917 * @actor: handler that splices the data
918 *
919 * Description:
920 * This function does little more than loop over the pipe and call
921 * @actor to do the actual moving of a single struct pipe_buffer to
922 * the desired destination. See pipe_to_file, pipe_to_sendpage, or
923 * pipe_to_user.
924 *
Jens Axboe83f91352006-04-02 23:05:09 +0200925 */
Jens Axboec66ab6f2007-06-12 21:17:17 +0200926ssize_t __splice_from_pipe(struct pipe_inode_info *pipe, struct splice_desc *sd,
927 splice_actor *actor)
Jens Axboe5274f052006-03-30 15:15:30 +0200928{
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200929 int ret;
Jens Axboe5274f052006-03-30 15:15:30 +0200930
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200931 splice_from_pipe_begin(sd);
932 do {
933 ret = splice_from_pipe_next(pipe, sd);
934 if (ret > 0)
935 ret = splice_from_pipe_feed(pipe, sd, actor);
936 } while (ret > 0);
937 splice_from_pipe_end(pipe, sd);
Jens Axboe5274f052006-03-30 15:15:30 +0200938
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200939 return sd->num_spliced ? sd->num_spliced : ret;
Jens Axboe5274f052006-03-30 15:15:30 +0200940}
Mark Fasheh40bee44e2007-03-21 13:11:02 +0100941EXPORT_SYMBOL(__splice_from_pipe);
Jens Axboe5274f052006-03-30 15:15:30 +0200942
Jens Axboe932cc6d2007-06-21 13:10:21 +0200943/**
944 * splice_from_pipe - splice data from a pipe to a file
945 * @pipe: pipe to splice from
946 * @out: file to splice to
947 * @ppos: position in @out
948 * @len: how many bytes to splice
949 * @flags: splice modifier flags
950 * @actor: handler that splices the data
951 *
952 * Description:
Miklos Szeredi29339702009-04-14 19:48:37 +0200953 * See __splice_from_pipe. This function locks the pipe inode,
Jens Axboe932cc6d2007-06-21 13:10:21 +0200954 * otherwise it's identical to __splice_from_pipe().
955 *
956 */
Mark Fasheh6da61802006-10-17 18:43:07 +0200957ssize_t splice_from_pipe(struct pipe_inode_info *pipe, struct file *out,
958 loff_t *ppos, size_t len, unsigned int flags,
959 splice_actor *actor)
960{
961 ssize_t ret;
Jens Axboec66ab6f2007-06-12 21:17:17 +0200962 struct splice_desc sd = {
963 .total_len = len,
964 .flags = flags,
965 .pos = *ppos,
Jens Axboe6a14b902007-06-14 13:08:55 +0200966 .u.file = out,
Jens Axboec66ab6f2007-06-12 21:17:17 +0200967 };
Mark Fasheh6da61802006-10-17 18:43:07 +0200968
Miklos Szeredi61e0d472009-04-14 19:48:41 +0200969 pipe_lock(pipe);
Jens Axboec66ab6f2007-06-12 21:17:17 +0200970 ret = __splice_from_pipe(pipe, &sd, actor);
Miklos Szeredi61e0d472009-04-14 19:48:41 +0200971 pipe_unlock(pipe);
Mark Fasheh6da61802006-10-17 18:43:07 +0200972
973 return ret;
974}
975
976/**
Jens Axboe83f91352006-04-02 23:05:09 +0200977 * generic_file_splice_write - splice data from a pipe to a file
Ingo Molnar3a326a22006-04-10 15:18:35 +0200978 * @pipe: pipe info
Jens Axboe83f91352006-04-02 23:05:09 +0200979 * @out: file to write to
Jens Axboe932cc6d2007-06-21 13:10:21 +0200980 * @ppos: position in @out
Jens Axboe83f91352006-04-02 23:05:09 +0200981 * @len: number of bytes to splice
982 * @flags: splice modifier flags
983 *
Jens Axboe932cc6d2007-06-21 13:10:21 +0200984 * Description:
985 * Will either move or copy pages (determined by @flags options) from
986 * the given pipe inode to the given file.
Jens Axboe83f91352006-04-02 23:05:09 +0200987 *
988 */
Ingo Molnar3a326a22006-04-10 15:18:35 +0200989ssize_t
990generic_file_splice_write(struct pipe_inode_info *pipe, struct file *out,
Jens Axboecbb7e572006-04-11 14:57:50 +0200991 loff_t *ppos, size_t len, unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +0200992{
Jens Axboe4f6f0bd2006-04-02 23:04:46 +0200993 struct address_space *mapping = out->f_mapping;
Jens Axboe8c34e2d2006-10-17 19:43:22 +0200994 struct inode *inode = mapping->host;
Miklos Szeredi7f3d4ee2008-05-07 09:22:39 +0200995 struct splice_desc sd = {
996 .total_len = len,
997 .flags = flags,
998 .pos = *ppos,
999 .u.file = out,
1000 };
Ingo Molnar3a326a22006-04-10 15:18:35 +02001001 ssize_t ret;
Jens Axboe8c34e2d2006-10-17 19:43:22 +02001002
Jan Kara14da9202012-06-12 16:20:37 +02001003 sb_start_write(inode->i_sb);
1004
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001005 pipe_lock(pipe);
Miklos Szeredieb443e52009-04-14 19:48:38 +02001006
1007 splice_from_pipe_begin(&sd);
1008 do {
1009 ret = splice_from_pipe_next(pipe, &sd);
1010 if (ret <= 0)
1011 break;
1012
1013 mutex_lock_nested(&inode->i_mutex, I_MUTEX_CHILD);
1014 ret = file_remove_suid(out);
Miklos Szeredi723590e2009-08-15 08:43:22 +02001015 if (!ret) {
Josef Bacikc3b2da32012-03-26 09:59:21 -04001016 ret = file_update_time(out);
1017 if (!ret)
1018 ret = splice_from_pipe_feed(pipe, &sd,
1019 pipe_to_file);
Miklos Szeredi723590e2009-08-15 08:43:22 +02001020 }
Miklos Szeredieb443e52009-04-14 19:48:38 +02001021 mutex_unlock(&inode->i_mutex);
1022 } while (ret > 0);
1023 splice_from_pipe_end(pipe, &sd);
1024
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001025 pipe_unlock(pipe);
Miklos Szeredieb443e52009-04-14 19:48:38 +02001026
1027 if (sd.num_spliced)
1028 ret = sd.num_spliced;
1029
Jens Axboea4514eb2006-04-19 15:57:05 +02001030 if (ret > 0) {
Jan Kara148f9482009-08-17 19:52:36 +02001031 int err;
Jens Axboe17ee4f42007-06-15 13:10:37 +02001032
Jan Kara148f9482009-08-17 19:52:36 +02001033 err = generic_write_sync(out, *ppos, ret);
1034 if (err)
1035 ret = err;
1036 else
1037 *ppos += ret;
Namjae Jeond0e1d662012-12-11 16:00:21 -08001038 balance_dirty_pages_ratelimited(mapping);
Jens Axboe4f6f0bd2006-04-02 23:04:46 +02001039 }
Jan Kara14da9202012-06-12 16:20:37 +02001040 sb_end_write(inode->i_sb);
Jens Axboe4f6f0bd2006-04-02 23:04:46 +02001041
1042 return ret;
Jens Axboe5274f052006-03-30 15:15:30 +02001043}
1044
Jens Axboe059a8f32006-04-02 23:06:05 +02001045EXPORT_SYMBOL(generic_file_splice_write);
1046
Miklos Szeredib2858d72009-05-19 11:37:46 +02001047static int write_pipe_buf(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
1048 struct splice_desc *sd)
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +02001049{
Miklos Szeredib2858d72009-05-19 11:37:46 +02001050 int ret;
1051 void *data;
Al Viro06ae43f2013-03-20 13:19:30 -04001052 loff_t tmp = sd->pos;
Miklos Szeredib2858d72009-05-19 11:37:46 +02001053
Miklos Szeredib2858d72009-05-19 11:37:46 +02001054 data = buf->ops->map(pipe, buf, 0);
Al Viro06ae43f2013-03-20 13:19:30 -04001055 ret = __kernel_write(sd->u.file, data + buf->offset, sd->len, &tmp);
Miklos Szeredib2858d72009-05-19 11:37:46 +02001056 buf->ops->unmap(pipe, buf, data);
1057
1058 return ret;
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +02001059}
1060
1061static ssize_t default_file_splice_write(struct pipe_inode_info *pipe,
1062 struct file *out, loff_t *ppos,
1063 size_t len, unsigned int flags)
1064{
Miklos Szeredib2858d72009-05-19 11:37:46 +02001065 ssize_t ret;
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +02001066
Miklos Szeredib2858d72009-05-19 11:37:46 +02001067 ret = splice_from_pipe(pipe, out, ppos, len, flags, write_pipe_buf);
1068 if (ret > 0)
1069 *ppos += ret;
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +02001070
Miklos Szeredib2858d72009-05-19 11:37:46 +02001071 return ret;
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +02001072}
1073
Jens Axboe83f91352006-04-02 23:05:09 +02001074/**
1075 * generic_splice_sendpage - splice data from a pipe to a socket
Jens Axboe932cc6d2007-06-21 13:10:21 +02001076 * @pipe: pipe to splice from
Jens Axboe83f91352006-04-02 23:05:09 +02001077 * @out: socket to write to
Jens Axboe932cc6d2007-06-21 13:10:21 +02001078 * @ppos: position in @out
Jens Axboe83f91352006-04-02 23:05:09 +02001079 * @len: number of bytes to splice
1080 * @flags: splice modifier flags
1081 *
Jens Axboe932cc6d2007-06-21 13:10:21 +02001082 * Description:
1083 * Will send @len bytes from the pipe to a network socket. No data copying
1084 * is involved.
Jens Axboe83f91352006-04-02 23:05:09 +02001085 *
1086 */
Ingo Molnar3a326a22006-04-10 15:18:35 +02001087ssize_t generic_splice_sendpage(struct pipe_inode_info *pipe, struct file *out,
Jens Axboecbb7e572006-04-11 14:57:50 +02001088 loff_t *ppos, size_t len, unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +02001089{
Jens Axboe00522fb2006-04-26 14:39:29 +02001090 return splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_sendpage);
Jens Axboe5274f052006-03-30 15:15:30 +02001091}
1092
Jens Axboe059a8f32006-04-02 23:06:05 +02001093EXPORT_SYMBOL(generic_splice_sendpage);
Jeff Garzika0f06782006-03-30 23:06:13 -05001094
Jens Axboe83f91352006-04-02 23:05:09 +02001095/*
1096 * Attempt to initiate a splice from pipe to file.
1097 */
Ingo Molnar3a326a22006-04-10 15:18:35 +02001098static long do_splice_from(struct pipe_inode_info *pipe, struct file *out,
Jens Axboecbb7e572006-04-11 14:57:50 +02001099 loff_t *ppos, size_t len, unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +02001100{
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +02001101 ssize_t (*splice_write)(struct pipe_inode_info *, struct file *,
1102 loff_t *, size_t, unsigned int);
Jens Axboe5274f052006-03-30 15:15:30 +02001103 int ret;
1104
Jens Axboe49570e92006-04-11 13:56:09 +02001105 if (unlikely(!(out->f_mode & FMODE_WRITE)))
Jens Axboe5274f052006-03-30 15:15:30 +02001106 return -EBADF;
1107
Linus Torvaldsefc968d2008-10-09 14:04:54 -07001108 if (unlikely(out->f_flags & O_APPEND))
1109 return -EINVAL;
1110
Jens Axboecbb7e572006-04-11 14:57:50 +02001111 ret = rw_verify_area(WRITE, out, ppos, len);
Jens Axboe5274f052006-03-30 15:15:30 +02001112 if (unlikely(ret < 0))
1113 return ret;
1114
Changli Gaocc56f7d2009-11-04 09:09:52 +01001115 if (out->f_op && out->f_op->splice_write)
1116 splice_write = out->f_op->splice_write;
1117 else
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +02001118 splice_write = default_file_splice_write;
1119
1120 return splice_write(pipe, out, ppos, len, flags);
Jens Axboe5274f052006-03-30 15:15:30 +02001121}
1122
Jens Axboe83f91352006-04-02 23:05:09 +02001123/*
1124 * Attempt to initiate a splice from a file to a pipe.
1125 */
Jens Axboecbb7e572006-04-11 14:57:50 +02001126static long do_splice_to(struct file *in, loff_t *ppos,
1127 struct pipe_inode_info *pipe, size_t len,
1128 unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +02001129{
Miklos Szeredi68181732009-05-07 15:37:36 +02001130 ssize_t (*splice_read)(struct file *, loff_t *,
1131 struct pipe_inode_info *, size_t, unsigned int);
Jens Axboe5274f052006-03-30 15:15:30 +02001132 int ret;
1133
Jens Axboe49570e92006-04-11 13:56:09 +02001134 if (unlikely(!(in->f_mode & FMODE_READ)))
Jens Axboe5274f052006-03-30 15:15:30 +02001135 return -EBADF;
1136
Jens Axboecbb7e572006-04-11 14:57:50 +02001137 ret = rw_verify_area(READ, in, ppos, len);
Jens Axboe5274f052006-03-30 15:15:30 +02001138 if (unlikely(ret < 0))
1139 return ret;
1140
Changli Gaocc56f7d2009-11-04 09:09:52 +01001141 if (in->f_op && in->f_op->splice_read)
1142 splice_read = in->f_op->splice_read;
1143 else
Miklos Szeredi68181732009-05-07 15:37:36 +02001144 splice_read = default_file_splice_read;
1145
1146 return splice_read(in, ppos, pipe, len, flags);
Jens Axboe5274f052006-03-30 15:15:30 +02001147}
1148
Jens Axboe932cc6d2007-06-21 13:10:21 +02001149/**
1150 * splice_direct_to_actor - splices data directly between two non-pipes
1151 * @in: file to splice from
1152 * @sd: actor information on where to splice to
1153 * @actor: handles the data splicing
1154 *
1155 * Description:
1156 * This is a special case helper to splice directly between two
1157 * points, without requiring an explicit pipe. Internally an allocated
Randy Dunlap79685b82007-07-27 08:08:51 +02001158 * pipe is cached in the process, and reused during the lifetime of
Jens Axboe932cc6d2007-06-21 13:10:21 +02001159 * that process.
1160 *
Jens Axboec66ab6f2007-06-12 21:17:17 +02001161 */
1162ssize_t splice_direct_to_actor(struct file *in, struct splice_desc *sd,
1163 splice_direct_actor *actor)
Jens Axboeb92ce552006-04-11 13:52:07 +02001164{
1165 struct pipe_inode_info *pipe;
1166 long ret, bytes;
1167 umode_t i_mode;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001168 size_t len;
1169 int i, flags;
Jens Axboeb92ce552006-04-11 13:52:07 +02001170
1171 /*
1172 * We require the input being a regular file, as we don't want to
1173 * randomly drop data for eg socket -> socket splicing. Use the
1174 * piped splicing for that!
1175 */
Al Viro496ad9a2013-01-23 17:07:38 -05001176 i_mode = file_inode(in)->i_mode;
Jens Axboeb92ce552006-04-11 13:52:07 +02001177 if (unlikely(!S_ISREG(i_mode) && !S_ISBLK(i_mode)))
1178 return -EINVAL;
1179
1180 /*
1181 * neither in nor out is a pipe, setup an internal pipe attached to
1182 * 'out' and transfer the wanted data from 'in' to 'out' through that
1183 */
1184 pipe = current->splice_pipe;
Jens Axboe49570e92006-04-11 13:56:09 +02001185 if (unlikely(!pipe)) {
Jens Axboeb92ce552006-04-11 13:52:07 +02001186 pipe = alloc_pipe_info(NULL);
1187 if (!pipe)
1188 return -ENOMEM;
1189
1190 /*
1191 * We don't have an immediate reader, but we'll read the stuff
Jens Axboe00522fb2006-04-26 14:39:29 +02001192 * out of the pipe right after the splice_to_pipe(). So set
Jens Axboeb92ce552006-04-11 13:52:07 +02001193 * PIPE_READERS appropriately.
1194 */
1195 pipe->readers = 1;
1196
1197 current->splice_pipe = pipe;
1198 }
1199
1200 /*
Ingo Molnar73d62d82006-04-11 13:57:21 +02001201 * Do the splice.
Jens Axboeb92ce552006-04-11 13:52:07 +02001202 */
1203 ret = 0;
1204 bytes = 0;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001205 len = sd->total_len;
1206 flags = sd->flags;
1207
1208 /*
1209 * Don't block on output, we have to drain the direct pipe.
1210 */
1211 sd->flags &= ~SPLICE_F_NONBLOCK;
Jens Axboeb92ce552006-04-11 13:52:07 +02001212
1213 while (len) {
Jens Axboe51a92c02007-07-13 14:11:43 +02001214 size_t read_len;
Tom Zanussia82c53a2008-05-09 13:28:36 +02001215 loff_t pos = sd->pos, prev_pos = pos;
Jens Axboeb92ce552006-04-11 13:52:07 +02001216
Jens Axboebcd4f3a2007-07-16 14:41:49 +02001217 ret = do_splice_to(in, &pos, pipe, len, flags);
Jens Axboe51a92c02007-07-13 14:11:43 +02001218 if (unlikely(ret <= 0))
Jens Axboeb92ce552006-04-11 13:52:07 +02001219 goto out_release;
1220
1221 read_len = ret;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001222 sd->total_len = read_len;
Jens Axboeb92ce552006-04-11 13:52:07 +02001223
1224 /*
1225 * NOTE: nonblocking mode only applies to the input. We
1226 * must not do the output in nonblocking mode as then we
1227 * could get stuck data in the internal pipe:
1228 */
Jens Axboec66ab6f2007-06-12 21:17:17 +02001229 ret = actor(pipe, sd);
Tom Zanussia82c53a2008-05-09 13:28:36 +02001230 if (unlikely(ret <= 0)) {
1231 sd->pos = prev_pos;
Jens Axboeb92ce552006-04-11 13:52:07 +02001232 goto out_release;
Tom Zanussia82c53a2008-05-09 13:28:36 +02001233 }
Jens Axboeb92ce552006-04-11 13:52:07 +02001234
1235 bytes += ret;
1236 len -= ret;
Jens Axboebcd4f3a2007-07-16 14:41:49 +02001237 sd->pos = pos;
Jens Axboeb92ce552006-04-11 13:52:07 +02001238
Tom Zanussia82c53a2008-05-09 13:28:36 +02001239 if (ret < read_len) {
1240 sd->pos = prev_pos + ret;
Jens Axboe51a92c02007-07-13 14:11:43 +02001241 goto out_release;
Tom Zanussia82c53a2008-05-09 13:28:36 +02001242 }
Jens Axboeb92ce552006-04-11 13:52:07 +02001243 }
1244
Jens Axboe9e971982008-01-29 21:05:57 +01001245done:
Jens Axboeb92ce552006-04-11 13:52:07 +02001246 pipe->nrbufs = pipe->curbuf = 0;
Jens Axboe80848702008-01-30 12:24:48 +01001247 file_accessed(in);
Jens Axboeb92ce552006-04-11 13:52:07 +02001248 return bytes;
1249
1250out_release:
1251 /*
1252 * If we did an incomplete transfer we must release
1253 * the pipe buffers in question:
1254 */
Jens Axboe35f3d142010-05-20 10:43:18 +02001255 for (i = 0; i < pipe->buffers; i++) {
Jens Axboeb92ce552006-04-11 13:52:07 +02001256 struct pipe_buffer *buf = pipe->bufs + i;
1257
1258 if (buf->ops) {
1259 buf->ops->release(pipe, buf);
1260 buf->ops = NULL;
1261 }
1262 }
Jens Axboeb92ce552006-04-11 13:52:07 +02001263
Jens Axboe9e971982008-01-29 21:05:57 +01001264 if (!bytes)
1265 bytes = ret;
Jens Axboeb92ce552006-04-11 13:52:07 +02001266
Jens Axboe9e971982008-01-29 21:05:57 +01001267 goto done;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001268}
1269EXPORT_SYMBOL(splice_direct_to_actor);
1270
1271static int direct_splice_actor(struct pipe_inode_info *pipe,
1272 struct splice_desc *sd)
1273{
Jens Axboe6a14b902007-06-14 13:08:55 +02001274 struct file *file = sd->u.file;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001275
Changli Gao2cb4b052010-06-29 13:09:18 +02001276 return do_splice_from(pipe, file, &file->f_pos, sd->total_len,
1277 sd->flags);
Jens Axboec66ab6f2007-06-12 21:17:17 +02001278}
1279
Jens Axboe932cc6d2007-06-21 13:10:21 +02001280/**
1281 * do_splice_direct - splices data directly between two files
1282 * @in: file to splice from
1283 * @ppos: input file offset
1284 * @out: file to splice to
1285 * @len: number of bytes to splice
1286 * @flags: splice modifier flags
1287 *
1288 * Description:
1289 * For use by do_sendfile(). splice can easily emulate sendfile, but
1290 * doing it in the application would incur an extra system call
1291 * (splice in + splice out, as compared to just sendfile()). So this helper
1292 * can splice directly through a process-private pipe.
1293 *
1294 */
Jens Axboec66ab6f2007-06-12 21:17:17 +02001295long do_splice_direct(struct file *in, loff_t *ppos, struct file *out,
1296 size_t len, unsigned int flags)
1297{
1298 struct splice_desc sd = {
1299 .len = len,
1300 .total_len = len,
1301 .flags = flags,
1302 .pos = *ppos,
Jens Axboe6a14b902007-06-14 13:08:55 +02001303 .u.file = out,
Jens Axboec66ab6f2007-06-12 21:17:17 +02001304 };
Jens Axboe51a92c02007-07-13 14:11:43 +02001305 long ret;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001306
1307 ret = splice_direct_to_actor(in, &sd, direct_splice_actor);
Jens Axboe51a92c02007-07-13 14:11:43 +02001308 if (ret > 0)
Tom Zanussia82c53a2008-05-09 13:28:36 +02001309 *ppos = sd.pos;
Jens Axboe51a92c02007-07-13 14:11:43 +02001310
Jens Axboec66ab6f2007-06-12 21:17:17 +02001311 return ret;
Jens Axboeb92ce552006-04-11 13:52:07 +02001312}
1313
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001314static int splice_pipe_to_pipe(struct pipe_inode_info *ipipe,
1315 struct pipe_inode_info *opipe,
1316 size_t len, unsigned int flags);
Jens Axboeddac0d32006-11-04 12:49:32 +01001317
1318/*
Jens Axboe83f91352006-04-02 23:05:09 +02001319 * Determine where to splice to/from.
1320 */
Ingo Molnar529565d2006-04-10 15:18:58 +02001321static long do_splice(struct file *in, loff_t __user *off_in,
1322 struct file *out, loff_t __user *off_out,
1323 size_t len, unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +02001324{
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001325 struct pipe_inode_info *ipipe;
1326 struct pipe_inode_info *opipe;
Jens Axboecbb7e572006-04-11 14:57:50 +02001327 loff_t offset, *off;
Jens Axboea4514eb2006-04-19 15:57:05 +02001328 long ret;
Jens Axboe5274f052006-03-30 15:15:30 +02001329
Linus Torvalds71993e62010-11-28 13:56:09 -08001330 ipipe = get_pipe_info(in);
1331 opipe = get_pipe_info(out);
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001332
1333 if (ipipe && opipe) {
1334 if (off_in || off_out)
1335 return -ESPIPE;
1336
1337 if (!(in->f_mode & FMODE_READ))
1338 return -EBADF;
1339
1340 if (!(out->f_mode & FMODE_WRITE))
1341 return -EBADF;
1342
1343 /* Splicing to self would be fun, but... */
1344 if (ipipe == opipe)
1345 return -EINVAL;
1346
1347 return splice_pipe_to_pipe(ipipe, opipe, len, flags);
1348 }
1349
1350 if (ipipe) {
Ingo Molnar529565d2006-04-10 15:18:58 +02001351 if (off_in)
1352 return -ESPIPE;
Jens Axboeb92ce552006-04-11 13:52:07 +02001353 if (off_out) {
Changli Gao19c9a492010-06-29 13:10:36 +02001354 if (!(out->f_mode & FMODE_PWRITE))
Jens Axboeb92ce552006-04-11 13:52:07 +02001355 return -EINVAL;
Jens Axboecbb7e572006-04-11 14:57:50 +02001356 if (copy_from_user(&offset, off_out, sizeof(loff_t)))
Jens Axboeb92ce552006-04-11 13:52:07 +02001357 return -EFAULT;
Jens Axboecbb7e572006-04-11 14:57:50 +02001358 off = &offset;
1359 } else
1360 off = &out->f_pos;
Ingo Molnar529565d2006-04-10 15:18:58 +02001361
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001362 ret = do_splice_from(ipipe, out, off, len, flags);
Jens Axboea4514eb2006-04-19 15:57:05 +02001363
1364 if (off_out && copy_to_user(off_out, off, sizeof(loff_t)))
1365 ret = -EFAULT;
1366
1367 return ret;
Ingo Molnar529565d2006-04-10 15:18:58 +02001368 }
Jens Axboe5274f052006-03-30 15:15:30 +02001369
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001370 if (opipe) {
Ingo Molnar529565d2006-04-10 15:18:58 +02001371 if (off_out)
1372 return -ESPIPE;
Jens Axboeb92ce552006-04-11 13:52:07 +02001373 if (off_in) {
Changli Gao19c9a492010-06-29 13:10:36 +02001374 if (!(in->f_mode & FMODE_PREAD))
Jens Axboeb92ce552006-04-11 13:52:07 +02001375 return -EINVAL;
Jens Axboecbb7e572006-04-11 14:57:50 +02001376 if (copy_from_user(&offset, off_in, sizeof(loff_t)))
Jens Axboeb92ce552006-04-11 13:52:07 +02001377 return -EFAULT;
Jens Axboecbb7e572006-04-11 14:57:50 +02001378 off = &offset;
1379 } else
1380 off = &in->f_pos;
Ingo Molnar529565d2006-04-10 15:18:58 +02001381
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001382 ret = do_splice_to(in, off, opipe, len, flags);
Jens Axboea4514eb2006-04-19 15:57:05 +02001383
1384 if (off_in && copy_to_user(off_in, off, sizeof(loff_t)))
1385 ret = -EFAULT;
1386
1387 return ret;
Ingo Molnar529565d2006-04-10 15:18:58 +02001388 }
Jens Axboe5274f052006-03-30 15:15:30 +02001389
1390 return -EINVAL;
1391}
1392
Jens Axboe912d35f2006-04-26 10:59:21 +02001393/*
1394 * Map an iov into an array of pages and offset/length tupples. With the
1395 * partial_page structure, we can map several non-contiguous ranges into
1396 * our ones pages[] map instead of splitting that operation into pieces.
1397 * Could easily be exported as a generic helper for other users, in which
1398 * case one would probably want to add a 'max_nr_pages' parameter as well.
1399 */
1400static int get_iovec_page_array(const struct iovec __user *iov,
1401 unsigned int nr_vecs, struct page **pages,
Eric Dumazetbd1a68b2012-04-04 11:08:51 +02001402 struct partial_page *partial, bool aligned,
Jens Axboe35f3d142010-05-20 10:43:18 +02001403 unsigned int pipe_buffers)
Jens Axboe912d35f2006-04-26 10:59:21 +02001404{
1405 int buffers = 0, error = 0;
1406
Jens Axboe912d35f2006-04-26 10:59:21 +02001407 while (nr_vecs) {
1408 unsigned long off, npages;
Linus Torvalds75723952007-10-01 13:17:28 -07001409 struct iovec entry;
Jens Axboe912d35f2006-04-26 10:59:21 +02001410 void __user *base;
1411 size_t len;
1412 int i;
1413
Linus Torvalds75723952007-10-01 13:17:28 -07001414 error = -EFAULT;
Nick Pigginbc40d732008-07-25 19:45:26 -07001415 if (copy_from_user(&entry, iov, sizeof(entry)))
Jens Axboe912d35f2006-04-26 10:59:21 +02001416 break;
Linus Torvalds75723952007-10-01 13:17:28 -07001417
1418 base = entry.iov_base;
1419 len = entry.iov_len;
Jens Axboe912d35f2006-04-26 10:59:21 +02001420
1421 /*
1422 * Sanity check this iovec. 0 read succeeds.
1423 */
Linus Torvalds75723952007-10-01 13:17:28 -07001424 error = 0;
Jens Axboe912d35f2006-04-26 10:59:21 +02001425 if (unlikely(!len))
1426 break;
1427 error = -EFAULT;
Bastian Blank712a30e2008-02-10 16:47:57 +02001428 if (!access_ok(VERIFY_READ, base, len))
Jens Axboe912d35f2006-04-26 10:59:21 +02001429 break;
1430
1431 /*
1432 * Get this base offset and number of pages, then map
1433 * in the user pages.
1434 */
1435 off = (unsigned long) base & ~PAGE_MASK;
Jens Axboe7afa6fd2006-05-01 20:02:33 +02001436
1437 /*
1438 * If asked for alignment, the offset must be zero and the
1439 * length a multiple of the PAGE_SIZE.
1440 */
1441 error = -EINVAL;
1442 if (aligned && (off || len & ~PAGE_MASK))
1443 break;
1444
Jens Axboe912d35f2006-04-26 10:59:21 +02001445 npages = (off + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
Jens Axboe35f3d142010-05-20 10:43:18 +02001446 if (npages > pipe_buffers - buffers)
1447 npages = pipe_buffers - buffers;
Jens Axboe912d35f2006-04-26 10:59:21 +02001448
Nick Pigginbc40d732008-07-25 19:45:26 -07001449 error = get_user_pages_fast((unsigned long)base, npages,
1450 0, &pages[buffers]);
Jens Axboe912d35f2006-04-26 10:59:21 +02001451
1452 if (unlikely(error <= 0))
1453 break;
1454
1455 /*
1456 * Fill this contiguous range into the partial page map.
1457 */
1458 for (i = 0; i < error; i++) {
Jens Axboe75914892006-05-02 12:57:18 +02001459 const int plen = min_t(size_t, len, PAGE_SIZE - off);
Jens Axboe912d35f2006-04-26 10:59:21 +02001460
1461 partial[buffers].offset = off;
1462 partial[buffers].len = plen;
1463
1464 off = 0;
1465 len -= plen;
1466 buffers++;
1467 }
1468
1469 /*
1470 * We didn't complete this iov, stop here since it probably
1471 * means we have to move some of this into a pipe to
1472 * be able to continue.
1473 */
1474 if (len)
1475 break;
1476
1477 /*
1478 * Don't continue if we mapped fewer pages than we asked for,
1479 * or if we mapped the max number of pages that we have
1480 * room for.
1481 */
Jens Axboe35f3d142010-05-20 10:43:18 +02001482 if (error < npages || buffers == pipe_buffers)
Jens Axboe912d35f2006-04-26 10:59:21 +02001483 break;
1484
1485 nr_vecs--;
1486 iov++;
1487 }
1488
Jens Axboe912d35f2006-04-26 10:59:21 +02001489 if (buffers)
1490 return buffers;
1491
1492 return error;
1493}
1494
Jens Axboe6a14b902007-06-14 13:08:55 +02001495static int pipe_to_user(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
1496 struct splice_desc *sd)
1497{
1498 char *src;
1499 int ret;
1500
Jens Axboe6a14b902007-06-14 13:08:55 +02001501 /*
1502 * See if we can use the atomic maps, by prefaulting in the
1503 * pages and doing an atomic copy
1504 */
1505 if (!fault_in_pages_writeable(sd->u.userptr, sd->len)) {
1506 src = buf->ops->map(pipe, buf, 1);
1507 ret = __copy_to_user_inatomic(sd->u.userptr, src + buf->offset,
1508 sd->len);
1509 buf->ops->unmap(pipe, buf, src);
1510 if (!ret) {
1511 ret = sd->len;
1512 goto out;
1513 }
1514 }
1515
1516 /*
1517 * No dice, use slow non-atomic map and copy
1518 */
1519 src = buf->ops->map(pipe, buf, 0);
1520
1521 ret = sd->len;
1522 if (copy_to_user(sd->u.userptr, src + buf->offset, sd->len))
1523 ret = -EFAULT;
1524
Jens Axboe6866bef2007-10-16 10:01:29 +02001525 buf->ops->unmap(pipe, buf, src);
Jens Axboe6a14b902007-06-14 13:08:55 +02001526out:
1527 if (ret > 0)
1528 sd->u.userptr += ret;
Jens Axboe6a14b902007-06-14 13:08:55 +02001529 return ret;
1530}
1531
1532/*
1533 * For lack of a better implementation, implement vmsplice() to userspace
1534 * as a simple copy of the pipes pages to the user iov.
1535 */
1536static long vmsplice_to_user(struct file *file, const struct iovec __user *iov,
1537 unsigned long nr_segs, unsigned int flags)
1538{
1539 struct pipe_inode_info *pipe;
1540 struct splice_desc sd;
1541 ssize_t size;
1542 int error;
1543 long ret;
1544
Linus Torvalds71993e62010-11-28 13:56:09 -08001545 pipe = get_pipe_info(file);
Jens Axboe6a14b902007-06-14 13:08:55 +02001546 if (!pipe)
1547 return -EBADF;
1548
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001549 pipe_lock(pipe);
Jens Axboe6a14b902007-06-14 13:08:55 +02001550
1551 error = ret = 0;
1552 while (nr_segs) {
1553 void __user *base;
1554 size_t len;
1555
1556 /*
1557 * Get user address base and length for this iovec.
1558 */
1559 error = get_user(base, &iov->iov_base);
1560 if (unlikely(error))
1561 break;
1562 error = get_user(len, &iov->iov_len);
1563 if (unlikely(error))
1564 break;
1565
1566 /*
1567 * Sanity check this iovec. 0 read succeeds.
1568 */
1569 if (unlikely(!len))
1570 break;
1571 if (unlikely(!base)) {
1572 error = -EFAULT;
1573 break;
1574 }
1575
Jens Axboe88119302008-02-08 08:49:14 -08001576 if (unlikely(!access_ok(VERIFY_WRITE, base, len))) {
1577 error = -EFAULT;
1578 break;
1579 }
1580
Jens Axboe6a14b902007-06-14 13:08:55 +02001581 sd.len = 0;
1582 sd.total_len = len;
1583 sd.flags = flags;
1584 sd.u.userptr = base;
1585 sd.pos = 0;
1586
1587 size = __splice_from_pipe(pipe, &sd, pipe_to_user);
1588 if (size < 0) {
1589 if (!ret)
1590 ret = size;
1591
1592 break;
1593 }
1594
1595 ret += size;
1596
1597 if (size < len)
1598 break;
1599
1600 nr_segs--;
1601 iov++;
1602 }
1603
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001604 pipe_unlock(pipe);
Jens Axboe6a14b902007-06-14 13:08:55 +02001605
1606 if (!ret)
1607 ret = error;
1608
1609 return ret;
1610}
1611
Jens Axboe912d35f2006-04-26 10:59:21 +02001612/*
1613 * vmsplice splices a user address range into a pipe. It can be thought of
1614 * as splice-from-memory, where the regular splice is splice-from-file (or
1615 * to file). In both cases the output is a pipe, naturally.
Jens Axboe912d35f2006-04-26 10:59:21 +02001616 */
Jens Axboe6a14b902007-06-14 13:08:55 +02001617static long vmsplice_to_pipe(struct file *file, const struct iovec __user *iov,
1618 unsigned long nr_segs, unsigned int flags)
Jens Axboe912d35f2006-04-26 10:59:21 +02001619{
Jens Axboeddac0d32006-11-04 12:49:32 +01001620 struct pipe_inode_info *pipe;
Jens Axboe35f3d142010-05-20 10:43:18 +02001621 struct page *pages[PIPE_DEF_BUFFERS];
1622 struct partial_page partial[PIPE_DEF_BUFFERS];
Jens Axboe912d35f2006-04-26 10:59:21 +02001623 struct splice_pipe_desc spd = {
1624 .pages = pages,
1625 .partial = partial,
Eric Dumazet047fe362012-06-12 15:24:40 +02001626 .nr_pages_max = PIPE_DEF_BUFFERS,
Jens Axboe912d35f2006-04-26 10:59:21 +02001627 .flags = flags,
1628 .ops = &user_page_pipe_buf_ops,
Jens Axboebbdfc2f2007-11-06 23:29:47 -08001629 .spd_release = spd_release_page,
Jens Axboe912d35f2006-04-26 10:59:21 +02001630 };
Jens Axboe35f3d142010-05-20 10:43:18 +02001631 long ret;
Jens Axboe912d35f2006-04-26 10:59:21 +02001632
Linus Torvalds71993e62010-11-28 13:56:09 -08001633 pipe = get_pipe_info(file);
Jens Axboeddac0d32006-11-04 12:49:32 +01001634 if (!pipe)
Jens Axboe912d35f2006-04-26 10:59:21 +02001635 return -EBADF;
Jens Axboe912d35f2006-04-26 10:59:21 +02001636
Jens Axboe35f3d142010-05-20 10:43:18 +02001637 if (splice_grow_spd(pipe, &spd))
1638 return -ENOMEM;
Jens Axboe912d35f2006-04-26 10:59:21 +02001639
Jens Axboe35f3d142010-05-20 10:43:18 +02001640 spd.nr_pages = get_iovec_page_array(iov, nr_segs, spd.pages,
Eric Dumazetbd1a68b2012-04-04 11:08:51 +02001641 spd.partial, false,
Eric Dumazet047fe362012-06-12 15:24:40 +02001642 spd.nr_pages_max);
Jens Axboe35f3d142010-05-20 10:43:18 +02001643 if (spd.nr_pages <= 0)
1644 ret = spd.nr_pages;
1645 else
1646 ret = splice_to_pipe(pipe, &spd);
1647
Eric Dumazet047fe362012-06-12 15:24:40 +02001648 splice_shrink_spd(&spd);
Jens Axboe35f3d142010-05-20 10:43:18 +02001649 return ret;
Jens Axboe912d35f2006-04-26 10:59:21 +02001650}
1651
Jens Axboe6a14b902007-06-14 13:08:55 +02001652/*
1653 * Note that vmsplice only really supports true splicing _from_ user memory
1654 * to a pipe, not the other way around. Splicing from user memory is a simple
1655 * operation that can be supported without any funky alignment restrictions
1656 * or nasty vm tricks. We simply map in the user memory and fill them into
1657 * a pipe. The reverse isn't quite as easy, though. There are two possible
1658 * solutions for that:
1659 *
1660 * - memcpy() the data internally, at which point we might as well just
1661 * do a regular read() on the buffer anyway.
1662 * - Lots of nasty vm tricks, that are neither fast nor flexible (it
1663 * has restriction limitations on both ends of the pipe).
1664 *
1665 * Currently we punt and implement it as a normal copy, see pipe_to_user().
1666 *
1667 */
Heiko Carstens836f92a2009-01-14 14:14:33 +01001668SYSCALL_DEFINE4(vmsplice, int, fd, const struct iovec __user *, iov,
1669 unsigned long, nr_segs, unsigned int, flags)
Jens Axboe912d35f2006-04-26 10:59:21 +02001670{
Al Viro2903ff02012-08-28 12:52:22 -04001671 struct fd f;
Jens Axboe912d35f2006-04-26 10:59:21 +02001672 long error;
Jens Axboe912d35f2006-04-26 10:59:21 +02001673
Jens Axboe6a14b902007-06-14 13:08:55 +02001674 if (unlikely(nr_segs > UIO_MAXIOV))
1675 return -EINVAL;
1676 else if (unlikely(!nr_segs))
1677 return 0;
1678
Jens Axboe912d35f2006-04-26 10:59:21 +02001679 error = -EBADF;
Al Viro2903ff02012-08-28 12:52:22 -04001680 f = fdget(fd);
1681 if (f.file) {
1682 if (f.file->f_mode & FMODE_WRITE)
1683 error = vmsplice_to_pipe(f.file, iov, nr_segs, flags);
1684 else if (f.file->f_mode & FMODE_READ)
1685 error = vmsplice_to_user(f.file, iov, nr_segs, flags);
Jens Axboe912d35f2006-04-26 10:59:21 +02001686
Al Viro2903ff02012-08-28 12:52:22 -04001687 fdput(f);
Jens Axboe912d35f2006-04-26 10:59:21 +02001688 }
1689
1690 return error;
1691}
1692
Heiko Carstens836f92a2009-01-14 14:14:33 +01001693SYSCALL_DEFINE6(splice, int, fd_in, loff_t __user *, off_in,
1694 int, fd_out, loff_t __user *, off_out,
1695 size_t, len, unsigned int, flags)
Jens Axboe5274f052006-03-30 15:15:30 +02001696{
Al Viro2903ff02012-08-28 12:52:22 -04001697 struct fd in, out;
Jens Axboe5274f052006-03-30 15:15:30 +02001698 long error;
Jens Axboe5274f052006-03-30 15:15:30 +02001699
1700 if (unlikely(!len))
1701 return 0;
1702
1703 error = -EBADF;
Al Viro2903ff02012-08-28 12:52:22 -04001704 in = fdget(fd_in);
1705 if (in.file) {
1706 if (in.file->f_mode & FMODE_READ) {
1707 out = fdget(fd_out);
1708 if (out.file) {
1709 if (out.file->f_mode & FMODE_WRITE)
1710 error = do_splice(in.file, off_in,
1711 out.file, off_out,
Ingo Molnar529565d2006-04-10 15:18:58 +02001712 len, flags);
Al Viro2903ff02012-08-28 12:52:22 -04001713 fdput(out);
Jens Axboe5274f052006-03-30 15:15:30 +02001714 }
1715 }
Al Viro2903ff02012-08-28 12:52:22 -04001716 fdput(in);
Jens Axboe5274f052006-03-30 15:15:30 +02001717 }
Jens Axboe5274f052006-03-30 15:15:30 +02001718 return error;
1719}
Jens Axboe70524492006-04-11 15:51:17 +02001720
1721/*
Jens Axboeaadd06e2006-07-10 11:00:01 +02001722 * Make sure there's data to read. Wait for input if we can, otherwise
1723 * return an appropriate error.
1724 */
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001725static int ipipe_prep(struct pipe_inode_info *pipe, unsigned int flags)
Jens Axboeaadd06e2006-07-10 11:00:01 +02001726{
1727 int ret;
1728
1729 /*
1730 * Check ->nrbufs without the inode lock first. This function
1731 * is speculative anyways, so missing one is ok.
1732 */
1733 if (pipe->nrbufs)
1734 return 0;
1735
1736 ret = 0;
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001737 pipe_lock(pipe);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001738
1739 while (!pipe->nrbufs) {
1740 if (signal_pending(current)) {
1741 ret = -ERESTARTSYS;
1742 break;
1743 }
1744 if (!pipe->writers)
1745 break;
1746 if (!pipe->waiting_writers) {
1747 if (flags & SPLICE_F_NONBLOCK) {
1748 ret = -EAGAIN;
1749 break;
1750 }
1751 }
1752 pipe_wait(pipe);
1753 }
1754
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001755 pipe_unlock(pipe);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001756 return ret;
1757}
1758
1759/*
1760 * Make sure there's writeable room. Wait for room if we can, otherwise
1761 * return an appropriate error.
1762 */
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001763static int opipe_prep(struct pipe_inode_info *pipe, unsigned int flags)
Jens Axboeaadd06e2006-07-10 11:00:01 +02001764{
1765 int ret;
1766
1767 /*
1768 * Check ->nrbufs without the inode lock first. This function
1769 * is speculative anyways, so missing one is ok.
1770 */
Jens Axboe35f3d142010-05-20 10:43:18 +02001771 if (pipe->nrbufs < pipe->buffers)
Jens Axboeaadd06e2006-07-10 11:00:01 +02001772 return 0;
1773
1774 ret = 0;
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001775 pipe_lock(pipe);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001776
Jens Axboe35f3d142010-05-20 10:43:18 +02001777 while (pipe->nrbufs >= pipe->buffers) {
Jens Axboeaadd06e2006-07-10 11:00:01 +02001778 if (!pipe->readers) {
1779 send_sig(SIGPIPE, current, 0);
1780 ret = -EPIPE;
1781 break;
1782 }
1783 if (flags & SPLICE_F_NONBLOCK) {
1784 ret = -EAGAIN;
1785 break;
1786 }
1787 if (signal_pending(current)) {
1788 ret = -ERESTARTSYS;
1789 break;
1790 }
1791 pipe->waiting_writers++;
1792 pipe_wait(pipe);
1793 pipe->waiting_writers--;
1794 }
1795
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001796 pipe_unlock(pipe);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001797 return ret;
1798}
1799
1800/*
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001801 * Splice contents of ipipe to opipe.
1802 */
1803static int splice_pipe_to_pipe(struct pipe_inode_info *ipipe,
1804 struct pipe_inode_info *opipe,
1805 size_t len, unsigned int flags)
1806{
1807 struct pipe_buffer *ibuf, *obuf;
1808 int ret = 0, nbuf;
1809 bool input_wakeup = false;
1810
1811
1812retry:
1813 ret = ipipe_prep(ipipe, flags);
1814 if (ret)
1815 return ret;
1816
1817 ret = opipe_prep(opipe, flags);
1818 if (ret)
1819 return ret;
1820
1821 /*
1822 * Potential ABBA deadlock, work around it by ordering lock
1823 * grabbing by pipe info address. Otherwise two different processes
1824 * could deadlock (one doing tee from A -> B, the other from B -> A).
1825 */
1826 pipe_double_lock(ipipe, opipe);
1827
1828 do {
1829 if (!opipe->readers) {
1830 send_sig(SIGPIPE, current, 0);
1831 if (!ret)
1832 ret = -EPIPE;
1833 break;
1834 }
1835
1836 if (!ipipe->nrbufs && !ipipe->writers)
1837 break;
1838
1839 /*
1840 * Cannot make any progress, because either the input
1841 * pipe is empty or the output pipe is full.
1842 */
Jens Axboe35f3d142010-05-20 10:43:18 +02001843 if (!ipipe->nrbufs || opipe->nrbufs >= opipe->buffers) {
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001844 /* Already processed some buffers, break */
1845 if (ret)
1846 break;
1847
1848 if (flags & SPLICE_F_NONBLOCK) {
1849 ret = -EAGAIN;
1850 break;
1851 }
1852
1853 /*
1854 * We raced with another reader/writer and haven't
1855 * managed to process any buffers. A zero return
1856 * value means EOF, so retry instead.
1857 */
1858 pipe_unlock(ipipe);
1859 pipe_unlock(opipe);
1860 goto retry;
1861 }
1862
1863 ibuf = ipipe->bufs + ipipe->curbuf;
Jens Axboe35f3d142010-05-20 10:43:18 +02001864 nbuf = (opipe->curbuf + opipe->nrbufs) & (opipe->buffers - 1);
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001865 obuf = opipe->bufs + nbuf;
1866
1867 if (len >= ibuf->len) {
1868 /*
1869 * Simply move the whole buffer from ipipe to opipe
1870 */
1871 *obuf = *ibuf;
1872 ibuf->ops = NULL;
1873 opipe->nrbufs++;
Jens Axboe35f3d142010-05-20 10:43:18 +02001874 ipipe->curbuf = (ipipe->curbuf + 1) & (ipipe->buffers - 1);
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001875 ipipe->nrbufs--;
1876 input_wakeup = true;
1877 } else {
1878 /*
1879 * Get a reference to this pipe buffer,
1880 * so we can copy the contents over.
1881 */
1882 ibuf->ops->get(ipipe, ibuf);
1883 *obuf = *ibuf;
1884
1885 /*
1886 * Don't inherit the gift flag, we need to
1887 * prevent multiple steals of this page.
1888 */
1889 obuf->flags &= ~PIPE_BUF_FLAG_GIFT;
1890
1891 obuf->len = len;
1892 opipe->nrbufs++;
1893 ibuf->offset += obuf->len;
1894 ibuf->len -= obuf->len;
1895 }
1896 ret += obuf->len;
1897 len -= obuf->len;
1898 } while (len);
1899
1900 pipe_unlock(ipipe);
1901 pipe_unlock(opipe);
1902
1903 /*
1904 * If we put data in the output pipe, wakeup any potential readers.
1905 */
Namhyung Kim825cdcb2011-05-23 19:58:53 +02001906 if (ret > 0)
1907 wakeup_pipe_readers(opipe);
1908
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001909 if (input_wakeup)
1910 wakeup_pipe_writers(ipipe);
1911
1912 return ret;
1913}
1914
1915/*
Jens Axboe70524492006-04-11 15:51:17 +02001916 * Link contents of ipipe to opipe.
1917 */
1918static int link_pipe(struct pipe_inode_info *ipipe,
1919 struct pipe_inode_info *opipe,
1920 size_t len, unsigned int flags)
1921{
1922 struct pipe_buffer *ibuf, *obuf;
Jens Axboeaadd06e2006-07-10 11:00:01 +02001923 int ret = 0, i = 0, nbuf;
Jens Axboe70524492006-04-11 15:51:17 +02001924
1925 /*
1926 * Potential ABBA deadlock, work around it by ordering lock
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001927 * grabbing by pipe info address. Otherwise two different processes
Jens Axboe70524492006-04-11 15:51:17 +02001928 * could deadlock (one doing tee from A -> B, the other from B -> A).
1929 */
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001930 pipe_double_lock(ipipe, opipe);
Jens Axboe70524492006-04-11 15:51:17 +02001931
Jens Axboeaadd06e2006-07-10 11:00:01 +02001932 do {
Jens Axboe70524492006-04-11 15:51:17 +02001933 if (!opipe->readers) {
1934 send_sig(SIGPIPE, current, 0);
1935 if (!ret)
1936 ret = -EPIPE;
1937 break;
1938 }
Jens Axboe70524492006-04-11 15:51:17 +02001939
1940 /*
Jens Axboeaadd06e2006-07-10 11:00:01 +02001941 * If we have iterated all input buffers or ran out of
1942 * output room, break.
Jens Axboe70524492006-04-11 15:51:17 +02001943 */
Jens Axboe35f3d142010-05-20 10:43:18 +02001944 if (i >= ipipe->nrbufs || opipe->nrbufs >= opipe->buffers)
Jens Axboe70524492006-04-11 15:51:17 +02001945 break;
Jens Axboeaadd06e2006-07-10 11:00:01 +02001946
Jens Axboe35f3d142010-05-20 10:43:18 +02001947 ibuf = ipipe->bufs + ((ipipe->curbuf + i) & (ipipe->buffers-1));
1948 nbuf = (opipe->curbuf + opipe->nrbufs) & (opipe->buffers - 1);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001949
Jens Axboe2a27250e2006-04-19 15:56:40 +02001950 /*
Jens Axboeaadd06e2006-07-10 11:00:01 +02001951 * Get a reference to this pipe buffer,
1952 * so we can copy the contents over.
Jens Axboe2a27250e2006-04-19 15:56:40 +02001953 */
Jens Axboeaadd06e2006-07-10 11:00:01 +02001954 ibuf->ops->get(ipipe, ibuf);
Jens Axboe70524492006-04-11 15:51:17 +02001955
Jens Axboeaadd06e2006-07-10 11:00:01 +02001956 obuf = opipe->bufs + nbuf;
1957 *obuf = *ibuf;
Jens Axboe70524492006-04-11 15:51:17 +02001958
Jens Axboeaadd06e2006-07-10 11:00:01 +02001959 /*
1960 * Don't inherit the gift flag, we need to
1961 * prevent multiple steals of this page.
1962 */
1963 obuf->flags &= ~PIPE_BUF_FLAG_GIFT;
1964
1965 if (obuf->len > len)
1966 obuf->len = len;
1967
1968 opipe->nrbufs++;
1969 ret += obuf->len;
1970 len -= obuf->len;
1971 i++;
1972 } while (len);
Jens Axboe70524492006-04-11 15:51:17 +02001973
Jens Axboe02cf01a2008-02-20 10:34:51 +01001974 /*
1975 * return EAGAIN if we have the potential of some data in the
1976 * future, otherwise just return 0
1977 */
1978 if (!ret && ipipe->waiting_writers && (flags & SPLICE_F_NONBLOCK))
1979 ret = -EAGAIN;
1980
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001981 pipe_unlock(ipipe);
1982 pipe_unlock(opipe);
Jens Axboe70524492006-04-11 15:51:17 +02001983
Jens Axboeaadd06e2006-07-10 11:00:01 +02001984 /*
1985 * If we put data in the output pipe, wakeup any potential readers.
1986 */
Namhyung Kim825cdcb2011-05-23 19:58:53 +02001987 if (ret > 0)
1988 wakeup_pipe_readers(opipe);
Jens Axboe70524492006-04-11 15:51:17 +02001989
1990 return ret;
1991}
1992
1993/*
1994 * This is a tee(1) implementation that works on pipes. It doesn't copy
1995 * any data, it simply references the 'in' pages on the 'out' pipe.
1996 * The 'flags' used are the SPLICE_F_* variants, currently the only
1997 * applicable one is SPLICE_F_NONBLOCK.
1998 */
1999static long do_tee(struct file *in, struct file *out, size_t len,
2000 unsigned int flags)
2001{
Linus Torvalds71993e62010-11-28 13:56:09 -08002002 struct pipe_inode_info *ipipe = get_pipe_info(in);
2003 struct pipe_inode_info *opipe = get_pipe_info(out);
Jens Axboeaadd06e2006-07-10 11:00:01 +02002004 int ret = -EINVAL;
Jens Axboe70524492006-04-11 15:51:17 +02002005
2006 /*
Jens Axboeaadd06e2006-07-10 11:00:01 +02002007 * Duplicate the contents of ipipe to opipe without actually
2008 * copying the data.
Jens Axboe70524492006-04-11 15:51:17 +02002009 */
Jens Axboeaadd06e2006-07-10 11:00:01 +02002010 if (ipipe && opipe && ipipe != opipe) {
2011 /*
2012 * Keep going, unless we encounter an error. The ipipe/opipe
2013 * ordering doesn't really matter.
2014 */
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02002015 ret = ipipe_prep(ipipe, flags);
Jens Axboeaadd06e2006-07-10 11:00:01 +02002016 if (!ret) {
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02002017 ret = opipe_prep(opipe, flags);
Jens Axboe02cf01a2008-02-20 10:34:51 +01002018 if (!ret)
Jens Axboeaadd06e2006-07-10 11:00:01 +02002019 ret = link_pipe(ipipe, opipe, len, flags);
Jens Axboeaadd06e2006-07-10 11:00:01 +02002020 }
2021 }
Jens Axboe70524492006-04-11 15:51:17 +02002022
Jens Axboeaadd06e2006-07-10 11:00:01 +02002023 return ret;
Jens Axboe70524492006-04-11 15:51:17 +02002024}
2025
Heiko Carstens836f92a2009-01-14 14:14:33 +01002026SYSCALL_DEFINE4(tee, int, fdin, int, fdout, size_t, len, unsigned int, flags)
Jens Axboe70524492006-04-11 15:51:17 +02002027{
Al Viro2903ff02012-08-28 12:52:22 -04002028 struct fd in;
2029 int error;
Jens Axboe70524492006-04-11 15:51:17 +02002030
2031 if (unlikely(!len))
2032 return 0;
2033
2034 error = -EBADF;
Al Viro2903ff02012-08-28 12:52:22 -04002035 in = fdget(fdin);
2036 if (in.file) {
2037 if (in.file->f_mode & FMODE_READ) {
2038 struct fd out = fdget(fdout);
2039 if (out.file) {
2040 if (out.file->f_mode & FMODE_WRITE)
2041 error = do_tee(in.file, out.file,
2042 len, flags);
2043 fdput(out);
Jens Axboe70524492006-04-11 15:51:17 +02002044 }
2045 }
Al Viro2903ff02012-08-28 12:52:22 -04002046 fdput(in);
Jens Axboe70524492006-04-11 15:51:17 +02002047 }
2048
2049 return error;
2050}