blob: e6b25598c8c413d66026f96e6d6d1b351e28b62c [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 Axboef84d7512006-05-01 19:59:03 +0200139 .map = generic_pipe_buf_map,
140 .unmap = generic_pipe_buf_unmap,
Jens Axboecac36bb02007-06-14 13:10:48 +0200141 .confirm = page_cache_pipe_buf_confirm,
Jens Axboe5274f052006-03-30 15:15:30 +0200142 .release = page_cache_pipe_buf_release,
Jens Axboe5abc97a2006-03-30 15:16:46 +0200143 .steal = page_cache_pipe_buf_steal,
Jens Axboef84d7512006-05-01 19:59:03 +0200144 .get = generic_pipe_buf_get,
Jens Axboe5274f052006-03-30 15:15:30 +0200145};
146
Jens Axboe912d35f2006-04-26 10:59:21 +0200147static int user_page_pipe_buf_steal(struct pipe_inode_info *pipe,
148 struct pipe_buffer *buf)
149{
Jens Axboe7afa6fd2006-05-01 20:02:33 +0200150 if (!(buf->flags & PIPE_BUF_FLAG_GIFT))
151 return 1;
152
Jens Axboe14328732006-05-03 10:35:26 +0200153 buf->flags |= PIPE_BUF_FLAG_LRU;
Jens Axboe330ab712006-05-02 15:29:57 +0200154 return generic_pipe_buf_steal(pipe, buf);
Jens Axboe912d35f2006-04-26 10:59:21 +0200155}
156
Eric Dumazetd4c3cca2006-12-13 00:34:04 -0800157static const struct pipe_buf_operations user_page_pipe_buf_ops = {
Jens Axboe912d35f2006-04-26 10:59:21 +0200158 .can_merge = 0,
Jens Axboef84d7512006-05-01 19:59:03 +0200159 .map = generic_pipe_buf_map,
160 .unmap = generic_pipe_buf_unmap,
Jens Axboecac36bb02007-06-14 13:10:48 +0200161 .confirm = generic_pipe_buf_confirm,
Jens Axboe912d35f2006-04-26 10:59:21 +0200162 .release = page_cache_pipe_buf_release,
163 .steal = user_page_pipe_buf_steal,
Jens Axboef84d7512006-05-01 19:59:03 +0200164 .get = generic_pipe_buf_get,
Jens Axboe912d35f2006-04-26 10:59:21 +0200165};
166
Namhyung Kim825cdcb2011-05-23 19:58:53 +0200167static void wakeup_pipe_readers(struct pipe_inode_info *pipe)
168{
169 smp_mb();
170 if (waitqueue_active(&pipe->wait))
171 wake_up_interruptible(&pipe->wait);
172 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
173}
174
Jens Axboe932cc6d2007-06-21 13:10:21 +0200175/**
176 * splice_to_pipe - fill passed data into a pipe
177 * @pipe: pipe to fill
178 * @spd: data to fill
179 *
180 * Description:
Randy Dunlap79685b82007-07-27 08:08:51 +0200181 * @spd contains a map of pages and len/offset tuples, along with
Jens Axboe932cc6d2007-06-21 13:10:21 +0200182 * the struct pipe_buf_operations associated with these pages. This
183 * function will link that data to the pipe.
184 *
Jens Axboe83f91352006-04-02 23:05:09 +0200185 */
Jens Axboed6b29d72007-06-04 09:59:47 +0200186ssize_t splice_to_pipe(struct pipe_inode_info *pipe,
187 struct splice_pipe_desc *spd)
Jens Axboe5274f052006-03-30 15:15:30 +0200188{
Jens Axboe00de00b2007-06-15 13:14:22 +0200189 unsigned int spd_pages = spd->nr_pages;
Jens Axboe912d35f2006-04-26 10:59:21 +0200190 int ret, do_wakeup, page_nr;
Jens Axboe5274f052006-03-30 15:15:30 +0200191
192 ret = 0;
193 do_wakeup = 0;
Jens Axboe912d35f2006-04-26 10:59:21 +0200194 page_nr = 0;
Jens Axboe5274f052006-03-30 15:15:30 +0200195
Miklos Szeredi61e0d472009-04-14 19:48:41 +0200196 pipe_lock(pipe);
Jens Axboe5274f052006-03-30 15:15:30 +0200197
Jens Axboe5274f052006-03-30 15:15:30 +0200198 for (;;) {
Ingo Molnar3a326a22006-04-10 15:18:35 +0200199 if (!pipe->readers) {
Jens Axboe5274f052006-03-30 15:15:30 +0200200 send_sig(SIGPIPE, current, 0);
201 if (!ret)
202 ret = -EPIPE;
203 break;
204 }
205
Jens Axboe35f3d142010-05-20 10:43:18 +0200206 if (pipe->nrbufs < pipe->buffers) {
207 int newbuf = (pipe->curbuf + pipe->nrbufs) & (pipe->buffers - 1);
Ingo Molnar3a326a22006-04-10 15:18:35 +0200208 struct pipe_buffer *buf = pipe->bufs + newbuf;
Jens Axboe5274f052006-03-30 15:15:30 +0200209
Jens Axboe912d35f2006-04-26 10:59:21 +0200210 buf->page = spd->pages[page_nr];
211 buf->offset = spd->partial[page_nr].offset;
212 buf->len = spd->partial[page_nr].len;
Jens Axboe497f9622007-06-11 12:00:45 +0200213 buf->private = spd->partial[page_nr].private;
Jens Axboe912d35f2006-04-26 10:59:21 +0200214 buf->ops = spd->ops;
Jens Axboe7afa6fd2006-05-01 20:02:33 +0200215 if (spd->flags & SPLICE_F_GIFT)
216 buf->flags |= PIPE_BUF_FLAG_GIFT;
217
Jens Axboe6f767b02006-04-11 13:53:56 +0200218 pipe->nrbufs++;
Jens Axboe912d35f2006-04-26 10:59:21 +0200219 page_nr++;
220 ret += buf->len;
221
Al Viro6447a3c2013-03-21 11:01:38 -0400222 if (pipe->files)
Jens Axboe6f767b02006-04-11 13:53:56 +0200223 do_wakeup = 1;
Jens Axboe5274f052006-03-30 15:15:30 +0200224
Jens Axboe912d35f2006-04-26 10:59:21 +0200225 if (!--spd->nr_pages)
Jens Axboe5274f052006-03-30 15:15:30 +0200226 break;
Jens Axboe35f3d142010-05-20 10:43:18 +0200227 if (pipe->nrbufs < pipe->buffers)
Jens Axboe5274f052006-03-30 15:15:30 +0200228 continue;
229
230 break;
231 }
232
Jens Axboe912d35f2006-04-26 10:59:21 +0200233 if (spd->flags & SPLICE_F_NONBLOCK) {
Linus Torvalds29e35092006-04-02 12:46:35 -0700234 if (!ret)
235 ret = -EAGAIN;
236 break;
237 }
238
Jens Axboe5274f052006-03-30 15:15:30 +0200239 if (signal_pending(current)) {
240 if (!ret)
241 ret = -ERESTARTSYS;
242 break;
243 }
244
245 if (do_wakeup) {
Jens Axboec0bd1f62006-04-10 09:03:32 +0200246 smp_mb();
Ingo Molnar3a326a22006-04-10 15:18:35 +0200247 if (waitqueue_active(&pipe->wait))
248 wake_up_interruptible_sync(&pipe->wait);
249 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
Jens Axboe5274f052006-03-30 15:15:30 +0200250 do_wakeup = 0;
251 }
252
Ingo Molnar3a326a22006-04-10 15:18:35 +0200253 pipe->waiting_writers++;
254 pipe_wait(pipe);
255 pipe->waiting_writers--;
Jens Axboe5274f052006-03-30 15:15:30 +0200256 }
257
Miklos Szeredi61e0d472009-04-14 19:48:41 +0200258 pipe_unlock(pipe);
Jens Axboe5274f052006-03-30 15:15:30 +0200259
Namhyung Kim825cdcb2011-05-23 19:58:53 +0200260 if (do_wakeup)
261 wakeup_pipe_readers(pipe);
Jens Axboe5274f052006-03-30 15:15:30 +0200262
Jens Axboe00de00b2007-06-15 13:14:22 +0200263 while (page_nr < spd_pages)
Jens Axboebbdfc2f2007-11-06 23:29:47 -0800264 spd->spd_release(spd, page_nr++);
Jens Axboe5274f052006-03-30 15:15:30 +0200265
266 return ret;
267}
268
Hugh Dickins708e3502011-07-25 17:12:32 -0700269void spd_release_page(struct splice_pipe_desc *spd, unsigned int i)
Jens Axboebbdfc2f2007-11-06 23:29:47 -0800270{
271 page_cache_release(spd->pages[i]);
272}
273
Jens Axboe35f3d142010-05-20 10:43:18 +0200274/*
275 * Check if we need to grow the arrays holding pages and partial page
276 * descriptions.
277 */
Eric Dumazet047fe362012-06-12 15:24:40 +0200278int splice_grow_spd(const struct pipe_inode_info *pipe, struct splice_pipe_desc *spd)
Jens Axboe35f3d142010-05-20 10:43:18 +0200279{
Eric Dumazet047fe362012-06-12 15:24:40 +0200280 unsigned int buffers = ACCESS_ONCE(pipe->buffers);
281
282 spd->nr_pages_max = buffers;
283 if (buffers <= PIPE_DEF_BUFFERS)
Jens Axboe35f3d142010-05-20 10:43:18 +0200284 return 0;
285
Eric Dumazet047fe362012-06-12 15:24:40 +0200286 spd->pages = kmalloc(buffers * sizeof(struct page *), GFP_KERNEL);
287 spd->partial = kmalloc(buffers * sizeof(struct partial_page), GFP_KERNEL);
Jens Axboe35f3d142010-05-20 10:43:18 +0200288
289 if (spd->pages && spd->partial)
290 return 0;
291
292 kfree(spd->pages);
293 kfree(spd->partial);
294 return -ENOMEM;
295}
296
Eric Dumazet047fe362012-06-12 15:24:40 +0200297void splice_shrink_spd(struct splice_pipe_desc *spd)
Jens Axboe35f3d142010-05-20 10:43:18 +0200298{
Eric Dumazet047fe362012-06-12 15:24:40 +0200299 if (spd->nr_pages_max <= PIPE_DEF_BUFFERS)
Jens Axboe35f3d142010-05-20 10:43:18 +0200300 return;
301
302 kfree(spd->pages);
303 kfree(spd->partial);
304}
305
Ingo Molnar3a326a22006-04-10 15:18:35 +0200306static int
Jens Axboecbb7e572006-04-11 14:57:50 +0200307__generic_file_splice_read(struct file *in, loff_t *ppos,
308 struct pipe_inode_info *pipe, size_t len,
309 unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +0200310{
311 struct address_space *mapping = in->f_mapping;
Fengguang Wud8983912007-07-19 01:48:06 -0700312 unsigned int loff, nr_pages, req_pages;
Jens Axboe35f3d142010-05-20 10:43:18 +0200313 struct page *pages[PIPE_DEF_BUFFERS];
314 struct partial_page partial[PIPE_DEF_BUFFERS];
Jens Axboe5274f052006-03-30 15:15:30 +0200315 struct page *page;
Jens Axboe91ad66e2006-04-19 15:55:10 +0200316 pgoff_t index, end_index;
317 loff_t isize;
Jens Axboeeb207962006-04-27 11:05:22 +0200318 int error, page_nr;
Jens Axboe912d35f2006-04-26 10:59:21 +0200319 struct splice_pipe_desc spd = {
320 .pages = pages,
321 .partial = partial,
Eric Dumazet047fe362012-06-12 15:24:40 +0200322 .nr_pages_max = PIPE_DEF_BUFFERS,
Jens Axboe912d35f2006-04-26 10:59:21 +0200323 .flags = flags,
324 .ops = &page_cache_pipe_buf_ops,
Jens Axboebbdfc2f2007-11-06 23:29:47 -0800325 .spd_release = spd_release_page,
Jens Axboe912d35f2006-04-26 10:59:21 +0200326 };
Jens Axboe5274f052006-03-30 15:15:30 +0200327
Jens Axboe35f3d142010-05-20 10:43:18 +0200328 if (splice_grow_spd(pipe, &spd))
329 return -ENOMEM;
330
Jens Axboecbb7e572006-04-11 14:57:50 +0200331 index = *ppos >> PAGE_CACHE_SHIFT;
Jens Axboe912d35f2006-04-26 10:59:21 +0200332 loff = *ppos & ~PAGE_CACHE_MASK;
Fengguang Wud8983912007-07-19 01:48:06 -0700333 req_pages = (len + loff + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
Eric Dumazet047fe362012-06-12 15:24:40 +0200334 nr_pages = min(req_pages, spd.nr_pages_max);
Jens Axboe5274f052006-03-30 15:15:30 +0200335
336 /*
Jens Axboeeb207962006-04-27 11:05:22 +0200337 * Lookup the (hopefully) full range of pages we need.
338 */
Jens Axboe35f3d142010-05-20 10:43:18 +0200339 spd.nr_pages = find_get_pages_contig(mapping, index, nr_pages, spd.pages);
Fengguang Wu431a48202007-07-19 01:48:05 -0700340 index += spd.nr_pages;
Jens Axboeeb207962006-04-27 11:05:22 +0200341
342 /*
343 * If find_get_pages_contig() returned fewer pages than we needed,
Fengguang Wu431a48202007-07-19 01:48:05 -0700344 * readahead/allocate the rest and fill in the holes.
Jens Axboeeb207962006-04-27 11:05:22 +0200345 */
Fengguang Wu431a48202007-07-19 01:48:05 -0700346 if (spd.nr_pages < nr_pages)
Rusty Russellcf914a72007-07-19 01:48:08 -0700347 page_cache_sync_readahead(mapping, &in->f_ra, in,
348 index, req_pages - spd.nr_pages);
Fengguang Wu431a48202007-07-19 01:48:05 -0700349
Jens Axboe932cc6d2007-06-21 13:10:21 +0200350 error = 0;
Jens Axboeeb207962006-04-27 11:05:22 +0200351 while (spd.nr_pages < nr_pages) {
352 /*
353 * Page could be there, find_get_pages_contig() breaks on
354 * the first hole.
355 */
356 page = find_get_page(mapping, index);
357 if (!page) {
Jens Axboee27dedd2006-05-01 19:59:54 +0200358 /*
Jens Axboeeb207962006-04-27 11:05:22 +0200359 * page didn't exist, allocate one.
360 */
361 page = page_cache_alloc_cold(mapping);
362 if (!page)
363 break;
364
365 error = add_to_page_cache_lru(page, mapping, index,
Nick Piggin0ae0b5d2010-05-25 10:25:26 +0200366 GFP_KERNEL);
Jens Axboeeb207962006-04-27 11:05:22 +0200367 if (unlikely(error)) {
368 page_cache_release(page);
Jens Axboea0548872006-05-03 10:58:22 +0200369 if (error == -EEXIST)
370 continue;
Jens Axboeeb207962006-04-27 11:05:22 +0200371 break;
372 }
373 /*
374 * add_to_page_cache() locks the page, unlock it
375 * to avoid convoluting the logic below even more.
376 */
377 unlock_page(page);
378 }
379
Jens Axboe35f3d142010-05-20 10:43:18 +0200380 spd.pages[spd.nr_pages++] = page;
Jens Axboeeb207962006-04-27 11:05:22 +0200381 index++;
382 }
383
384 /*
385 * Now loop over the map and see if we need to start IO on any
386 * pages, fill in the partial map, etc.
387 */
388 index = *ppos >> PAGE_CACHE_SHIFT;
389 nr_pages = spd.nr_pages;
390 spd.nr_pages = 0;
391 for (page_nr = 0; page_nr < nr_pages; page_nr++) {
Jens Axboe82aa5d62006-04-20 13:05:48 +0200392 unsigned int this_len;
393
394 if (!len)
395 break;
396
397 /*
398 * this_len is the max we'll use from this page
399 */
Andrew Mortonba5f5d92006-04-25 15:33:34 +0200400 this_len = min_t(unsigned long, len, PAGE_CACHE_SIZE - loff);
Jens Axboe35f3d142010-05-20 10:43:18 +0200401 page = spd.pages[page_nr];
Jens Axboe7480a902006-04-11 13:52:47 +0200402
Fengguang Wua08a1662007-07-19 01:48:03 -0700403 if (PageReadahead(page))
Rusty Russellcf914a72007-07-19 01:48:08 -0700404 page_cache_async_readahead(mapping, &in->f_ra, in,
Fengguang Wud8983912007-07-19 01:48:06 -0700405 page, index, req_pages - page_nr);
Fengguang Wua08a1662007-07-19 01:48:03 -0700406
Jens Axboe7480a902006-04-11 13:52:47 +0200407 /*
408 * If the page isn't uptodate, we may need to start io on it
409 */
410 if (!PageUptodate(page)) {
Miklos Szeredi69650312010-08-03 12:48:50 +0200411 lock_page(page);
Jens Axboe7480a902006-04-11 13:52:47 +0200412
413 /*
Miklos Szeredi32502b82008-07-04 09:35:17 +0200414 * Page was truncated, or invalidated by the
415 * filesystem. Redo the find/create, but this time the
416 * page is kept locked, so there's no chance of another
417 * race with truncate/invalidate.
Jens Axboe7480a902006-04-11 13:52:47 +0200418 */
419 if (!page->mapping) {
420 unlock_page(page);
Miklos Szeredi32502b82008-07-04 09:35:17 +0200421 page = find_or_create_page(mapping, index,
422 mapping_gfp_mask(mapping));
423
424 if (!page) {
425 error = -ENOMEM;
426 break;
427 }
Jens Axboe35f3d142010-05-20 10:43:18 +0200428 page_cache_release(spd.pages[page_nr]);
429 spd.pages[page_nr] = page;
Jens Axboe7480a902006-04-11 13:52:47 +0200430 }
431 /*
432 * page was already under io and is now done, great
433 */
434 if (PageUptodate(page)) {
435 unlock_page(page);
436 goto fill_it;
437 }
438
Jens Axboe7480a902006-04-11 13:52:47 +0200439 /*
440 * need to read in the page
441 */
442 error = mapping->a_ops->readpage(in, page);
Jens Axboe7480a902006-04-11 13:52:47 +0200443 if (unlikely(error)) {
Jens Axboeeb207962006-04-27 11:05:22 +0200444 /*
445 * We really should re-lookup the page here,
446 * but it complicates things a lot. Instead
447 * lets just do what we already stored, and
448 * we'll get it the next time we are called.
449 */
Jens Axboe7480a902006-04-11 13:52:47 +0200450 if (error == AOP_TRUNCATED_PAGE)
Jens Axboeeb207962006-04-27 11:05:22 +0200451 error = 0;
452
Jens Axboe7480a902006-04-11 13:52:47 +0200453 break;
454 }
Jens Axboe620a3242007-06-07 09:39:42 +0200455 }
456fill_it:
457 /*
458 * i_size must be checked after PageUptodate.
459 */
460 isize = i_size_read(mapping->host);
461 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
462 if (unlikely(!isize || index > end_index))
463 break;
464
465 /*
466 * if this is the last page, see if we need to shrink
467 * the length and stop
468 */
469 if (end_index == index) {
470 unsigned int plen;
Jens Axboe91ad66e2006-04-19 15:55:10 +0200471
472 /*
Jens Axboe620a3242007-06-07 09:39:42 +0200473 * max good bytes in this page
Jens Axboe91ad66e2006-04-19 15:55:10 +0200474 */
Jens Axboe620a3242007-06-07 09:39:42 +0200475 plen = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
476 if (plen <= loff)
Jens Axboe91ad66e2006-04-19 15:55:10 +0200477 break;
Jens Axboe91ad66e2006-04-19 15:55:10 +0200478
479 /*
Jens Axboe620a3242007-06-07 09:39:42 +0200480 * force quit after adding this page
Jens Axboe91ad66e2006-04-19 15:55:10 +0200481 */
Jens Axboe620a3242007-06-07 09:39:42 +0200482 this_len = min(this_len, plen - loff);
483 len = this_len;
Jens Axboe7480a902006-04-11 13:52:47 +0200484 }
Jens Axboe620a3242007-06-07 09:39:42 +0200485
Jens Axboe35f3d142010-05-20 10:43:18 +0200486 spd.partial[page_nr].offset = loff;
487 spd.partial[page_nr].len = this_len;
Jens Axboe82aa5d62006-04-20 13:05:48 +0200488 len -= this_len;
Jens Axboe91ad66e2006-04-19 15:55:10 +0200489 loff = 0;
Jens Axboeeb207962006-04-27 11:05:22 +0200490 spd.nr_pages++;
491 index++;
Jens Axboe5274f052006-03-30 15:15:30 +0200492 }
493
Jens Axboeeb207962006-04-27 11:05:22 +0200494 /*
Hugh Dickins475ecad2007-06-07 09:36:00 +0200495 * Release any pages at the end, if we quit early. 'page_nr' is how far
Jens Axboeeb207962006-04-27 11:05:22 +0200496 * we got, 'nr_pages' is how many pages are in the map.
497 */
498 while (page_nr < nr_pages)
Jens Axboe35f3d142010-05-20 10:43:18 +0200499 page_cache_release(spd.pages[page_nr++]);
Fengguang Wuf4e6b492007-10-16 01:24:33 -0700500 in->f_ra.prev_pos = (loff_t)index << PAGE_CACHE_SHIFT;
Jens Axboeeb207962006-04-27 11:05:22 +0200501
Jens Axboe912d35f2006-04-26 10:59:21 +0200502 if (spd.nr_pages)
Jens Axboe35f3d142010-05-20 10:43:18 +0200503 error = splice_to_pipe(pipe, &spd);
Jens Axboe5274f052006-03-30 15:15:30 +0200504
Eric Dumazet047fe362012-06-12 15:24:40 +0200505 splice_shrink_spd(&spd);
Jens Axboe7480a902006-04-11 13:52:47 +0200506 return error;
Jens Axboe5274f052006-03-30 15:15:30 +0200507}
508
Jens Axboe83f91352006-04-02 23:05:09 +0200509/**
510 * generic_file_splice_read - splice data from file to a pipe
511 * @in: file to splice from
Jens Axboe932cc6d2007-06-21 13:10:21 +0200512 * @ppos: position in @in
Jens Axboe83f91352006-04-02 23:05:09 +0200513 * @pipe: pipe to splice to
514 * @len: number of bytes to splice
515 * @flags: splice modifier flags
516 *
Jens Axboe932cc6d2007-06-21 13:10:21 +0200517 * Description:
518 * Will read pages from given file and fill them into a pipe. Can be
519 * used as long as the address_space operations for the source implements
520 * a readpage() hook.
521 *
Jens Axboe83f91352006-04-02 23:05:09 +0200522 */
Jens Axboecbb7e572006-04-11 14:57:50 +0200523ssize_t generic_file_splice_read(struct file *in, loff_t *ppos,
524 struct pipe_inode_info *pipe, size_t len,
525 unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +0200526{
Jens Axboed366d3982007-06-01 14:54:11 +0200527 loff_t isize, left;
Jens Axboe8191ecd2008-04-10 08:24:25 +0200528 int ret;
Jens Axboed366d3982007-06-01 14:54:11 +0200529
530 isize = i_size_read(in->f_mapping->host);
531 if (unlikely(*ppos >= isize))
532 return 0;
533
534 left = isize - *ppos;
535 if (unlikely(left < len))
536 len = left;
Jens Axboe5274f052006-03-30 15:15:30 +0200537
Jens Axboe8191ecd2008-04-10 08:24:25 +0200538 ret = __generic_file_splice_read(in, ppos, pipe, len, flags);
Miklos Szeredi723590e2009-08-15 08:43:22 +0200539 if (ret > 0) {
Jens Axboecbb7e572006-04-11 14:57:50 +0200540 *ppos += ret;
Miklos Szeredi723590e2009-08-15 08:43:22 +0200541 file_accessed(in);
542 }
Jens Axboe5274f052006-03-30 15:15:30 +0200543
544 return ret;
545}
Jens Axboe059a8f32006-04-02 23:06:05 +0200546EXPORT_SYMBOL(generic_file_splice_read);
547
Miklos Szeredi68181732009-05-07 15:37:36 +0200548static const struct pipe_buf_operations default_pipe_buf_ops = {
549 .can_merge = 0,
550 .map = generic_pipe_buf_map,
551 .unmap = generic_pipe_buf_unmap,
552 .confirm = generic_pipe_buf_confirm,
553 .release = generic_pipe_buf_release,
554 .steal = generic_pipe_buf_steal,
555 .get = generic_pipe_buf_get,
556};
557
558static ssize_t kernel_readv(struct file *file, const struct iovec *vec,
559 unsigned long vlen, loff_t offset)
560{
561 mm_segment_t old_fs;
562 loff_t pos = offset;
563 ssize_t res;
564
565 old_fs = get_fs();
566 set_fs(get_ds());
567 /* The cast to a user pointer is valid due to the set_fs() */
568 res = vfs_readv(file, (const struct iovec __user *)vec, vlen, &pos);
569 set_fs(old_fs);
570
571 return res;
572}
573
Al Viro7bb307e2013-02-23 14:51:48 -0500574ssize_t kernel_write(struct file *file, const char *buf, size_t count,
Miklos Szeredib2858d72009-05-19 11:37:46 +0200575 loff_t pos)
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +0200576{
577 mm_segment_t old_fs;
578 ssize_t res;
579
580 old_fs = get_fs();
581 set_fs(get_ds());
582 /* The cast to a user pointer is valid due to the set_fs() */
Al Viro7bb307e2013-02-23 14:51:48 -0500583 res = vfs_write(file, (__force const char __user *)buf, count, &pos);
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +0200584 set_fs(old_fs);
585
586 return res;
587}
Al Viro7bb307e2013-02-23 14:51:48 -0500588EXPORT_SYMBOL(kernel_write);
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +0200589
Miklos Szeredi68181732009-05-07 15:37:36 +0200590ssize_t default_file_splice_read(struct file *in, loff_t *ppos,
591 struct pipe_inode_info *pipe, size_t len,
592 unsigned int flags)
593{
594 unsigned int nr_pages;
595 unsigned int nr_freed;
596 size_t offset;
Jens Axboe35f3d142010-05-20 10:43:18 +0200597 struct page *pages[PIPE_DEF_BUFFERS];
598 struct partial_page partial[PIPE_DEF_BUFFERS];
599 struct iovec *vec, __vec[PIPE_DEF_BUFFERS];
Miklos Szeredi68181732009-05-07 15:37:36 +0200600 ssize_t res;
601 size_t this_len;
602 int error;
603 int i;
604 struct splice_pipe_desc spd = {
605 .pages = pages,
606 .partial = partial,
Eric Dumazet047fe362012-06-12 15:24:40 +0200607 .nr_pages_max = PIPE_DEF_BUFFERS,
Miklos Szeredi68181732009-05-07 15:37:36 +0200608 .flags = flags,
609 .ops = &default_pipe_buf_ops,
610 .spd_release = spd_release_page,
611 };
612
Jens Axboe35f3d142010-05-20 10:43:18 +0200613 if (splice_grow_spd(pipe, &spd))
614 return -ENOMEM;
615
616 res = -ENOMEM;
617 vec = __vec;
Eric Dumazet047fe362012-06-12 15:24:40 +0200618 if (spd.nr_pages_max > PIPE_DEF_BUFFERS) {
619 vec = kmalloc(spd.nr_pages_max * sizeof(struct iovec), GFP_KERNEL);
Jens Axboe35f3d142010-05-20 10:43:18 +0200620 if (!vec)
621 goto shrink_ret;
622 }
623
Miklos Szeredi68181732009-05-07 15:37:36 +0200624 offset = *ppos & ~PAGE_CACHE_MASK;
625 nr_pages = (len + offset + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
626
Eric Dumazet047fe362012-06-12 15:24:40 +0200627 for (i = 0; i < nr_pages && i < spd.nr_pages_max && len; i++) {
Miklos Szeredi68181732009-05-07 15:37:36 +0200628 struct page *page;
629
Jens Axboe4f231222009-05-13 08:35:35 +0200630 page = alloc_page(GFP_USER);
Miklos Szeredi68181732009-05-07 15:37:36 +0200631 error = -ENOMEM;
632 if (!page)
633 goto err;
634
635 this_len = min_t(size_t, len, PAGE_CACHE_SIZE - offset);
Jens Axboe4f231222009-05-13 08:35:35 +0200636 vec[i].iov_base = (void __user *) page_address(page);
Miklos Szeredi68181732009-05-07 15:37:36 +0200637 vec[i].iov_len = this_len;
Jens Axboe35f3d142010-05-20 10:43:18 +0200638 spd.pages[i] = page;
Miklos Szeredi68181732009-05-07 15:37:36 +0200639 spd.nr_pages++;
640 len -= this_len;
641 offset = 0;
642 }
643
644 res = kernel_readv(in, vec, spd.nr_pages, *ppos);
Andrew Morton77f6bf52009-05-14 09:49:44 +0200645 if (res < 0) {
646 error = res;
Miklos Szeredi68181732009-05-07 15:37:36 +0200647 goto err;
Andrew Morton77f6bf52009-05-14 09:49:44 +0200648 }
Miklos Szeredi68181732009-05-07 15:37:36 +0200649
650 error = 0;
651 if (!res)
652 goto err;
653
654 nr_freed = 0;
655 for (i = 0; i < spd.nr_pages; i++) {
Miklos Szeredi68181732009-05-07 15:37:36 +0200656 this_len = min_t(size_t, vec[i].iov_len, res);
Jens Axboe35f3d142010-05-20 10:43:18 +0200657 spd.partial[i].offset = 0;
658 spd.partial[i].len = this_len;
Miklos Szeredi68181732009-05-07 15:37:36 +0200659 if (!this_len) {
Jens Axboe35f3d142010-05-20 10:43:18 +0200660 __free_page(spd.pages[i]);
661 spd.pages[i] = NULL;
Miklos Szeredi68181732009-05-07 15:37:36 +0200662 nr_freed++;
663 }
664 res -= this_len;
665 }
666 spd.nr_pages -= nr_freed;
667
668 res = splice_to_pipe(pipe, &spd);
669 if (res > 0)
670 *ppos += res;
671
Jens Axboe35f3d142010-05-20 10:43:18 +0200672shrink_ret:
673 if (vec != __vec)
674 kfree(vec);
Eric Dumazet047fe362012-06-12 15:24:40 +0200675 splice_shrink_spd(&spd);
Miklos Szeredi68181732009-05-07 15:37:36 +0200676 return res;
677
678err:
Jens Axboe4f231222009-05-13 08:35:35 +0200679 for (i = 0; i < spd.nr_pages; i++)
Jens Axboe35f3d142010-05-20 10:43:18 +0200680 __free_page(spd.pages[i]);
Jens Axboe4f231222009-05-13 08:35:35 +0200681
Jens Axboe35f3d142010-05-20 10:43:18 +0200682 res = error;
683 goto shrink_ret;
Miklos Szeredi68181732009-05-07 15:37:36 +0200684}
685EXPORT_SYMBOL(default_file_splice_read);
686
Jens Axboe5274f052006-03-30 15:15:30 +0200687/*
Jens Axboe4f6f0bd2006-04-02 23:04:46 +0200688 * Send 'sd->len' bytes to socket from 'sd->file' at position 'sd->pos'
Jens Axboe016b6612006-04-25 15:42:00 +0200689 * using sendpage(). Return the number of bytes sent.
Jens Axboe5274f052006-03-30 15:15:30 +0200690 */
Jens Axboe76ad4d12006-05-03 10:41:33 +0200691static int pipe_to_sendpage(struct pipe_inode_info *pipe,
Jens Axboe5274f052006-03-30 15:15:30 +0200692 struct pipe_buffer *buf, struct splice_desc *sd)
693{
Jens Axboe6a14b902007-06-14 13:08:55 +0200694 struct file *file = sd->u.file;
Jens Axboe5274f052006-03-30 15:15:30 +0200695 loff_t pos = sd->pos;
Michał Mirosława8adbe32010-12-17 08:56:44 +0100696 int more;
Jens Axboe5274f052006-03-30 15:15:30 +0200697
Michał Mirosława8adbe32010-12-17 08:56:44 +0100698 if (!likely(file->f_op && file->f_op->sendpage))
699 return -EINVAL;
Jens Axboe5274f052006-03-30 15:15:30 +0200700
Eric Dumazet35f9c092012-04-05 03:05:35 +0000701 more = (sd->flags & SPLICE_F_MORE) ? MSG_MORE : 0;
Eric Dumazetae62ca72013-01-06 18:21:49 +0000702
703 if (sd->len < sd->total_len && pipe->nrbufs > 1)
Eric Dumazet35f9c092012-04-05 03:05:35 +0000704 more |= MSG_SENDPAGE_NOTLAST;
Eric Dumazetae62ca72013-01-06 18:21:49 +0000705
Michał Mirosława8adbe32010-12-17 08:56:44 +0100706 return file->f_op->sendpage(file, buf->page, buf->offset,
707 sd->len, &pos, more);
Jens Axboe5274f052006-03-30 15:15:30 +0200708}
709
710/*
711 * This is a little more tricky than the file -> pipe splicing. There are
712 * basically three cases:
713 *
714 * - Destination page already exists in the address space and there
715 * are users of it. For that case we have no other option that
716 * copying the data. Tough luck.
717 * - Destination page already exists in the address space, but there
718 * are no users of it. Make sure it's uptodate, then drop it. Fall
719 * through to last case.
720 * - Destination page does not exist, we can add the pipe page to
721 * the page cache and avoid the copy.
722 *
Jens Axboe83f91352006-04-02 23:05:09 +0200723 * If asked to move pages to the output file (SPLICE_F_MOVE is set in
724 * sd->flags), we attempt to migrate pages from the pipe to the output
725 * file address space page cache. This is possible if no one else has
726 * the pipe page referenced outside of the pipe and page cache. If
727 * SPLICE_F_MOVE isn't set, or we cannot move the page, we simply create
728 * a new page in the output file page cache and fill/dirty that.
Jens Axboe5274f052006-03-30 15:15:30 +0200729 */
Miklos Szeredi328eaab2009-04-14 19:48:39 +0200730int pipe_to_file(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
731 struct splice_desc *sd)
Jens Axboe5274f052006-03-30 15:15:30 +0200732{
Jens Axboe6a14b902007-06-14 13:08:55 +0200733 struct file *file = sd->u.file;
Jens Axboe5274f052006-03-30 15:15:30 +0200734 struct address_space *mapping = file->f_mapping;
Jens Axboe016b6612006-04-25 15:42:00 +0200735 unsigned int offset, this_len;
Jens Axboe5274f052006-03-30 15:15:30 +0200736 struct page *page;
Nick Pigginafddba42007-10-16 01:25:01 -0700737 void *fsdata;
Jens Axboe3e7ee3e2006-04-02 23:11:04 +0200738 int ret;
Jens Axboe5274f052006-03-30 15:15:30 +0200739
Jens Axboe5274f052006-03-30 15:15:30 +0200740 offset = sd->pos & ~PAGE_CACHE_MASK;
741
Jens Axboe016b6612006-04-25 15:42:00 +0200742 this_len = sd->len;
743 if (this_len + offset > PAGE_CACHE_SIZE)
744 this_len = PAGE_CACHE_SIZE - offset;
745
Nick Pigginafddba42007-10-16 01:25:01 -0700746 ret = pagecache_write_begin(file, mapping, sd->pos, this_len,
747 AOP_FLAG_UNINTERRUPTIBLE, &page, &fsdata);
748 if (unlikely(ret))
749 goto out;
Jens Axboe5274f052006-03-30 15:15:30 +0200750
Jens Axboe0568b402006-05-01 19:50:48 +0200751 if (buf->page != page) {
Jens Axboe76ad4d12006-05-03 10:41:33 +0200752 char *src = buf->ops->map(pipe, buf, 1);
Cong Wange8e3c3d2011-11-25 23:14:27 +0800753 char *dst = kmap_atomic(page);
Jens Axboe5abc97a2006-03-30 15:16:46 +0200754
Jens Axboe016b6612006-04-25 15:42:00 +0200755 memcpy(dst + offset, src + buf->offset, this_len);
Jens Axboe5abc97a2006-03-30 15:16:46 +0200756 flush_dcache_page(page);
Cong Wange8e3c3d2011-11-25 23:14:27 +0800757 kunmap_atomic(dst);
Jens Axboe76ad4d12006-05-03 10:41:33 +0200758 buf->ops->unmap(pipe, buf, src);
Jens Axboe5abc97a2006-03-30 15:16:46 +0200759 }
Nick Pigginafddba42007-10-16 01:25:01 -0700760 ret = pagecache_write_end(file, mapping, sd->pos, this_len, this_len,
761 page, fsdata);
Jens Axboe5274f052006-03-30 15:15:30 +0200762out:
Jens Axboe5274f052006-03-30 15:15:30 +0200763 return ret;
764}
Miklos Szeredi328eaab2009-04-14 19:48:39 +0200765EXPORT_SYMBOL(pipe_to_file);
Jens Axboe5274f052006-03-30 15:15:30 +0200766
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200767static void wakeup_pipe_writers(struct pipe_inode_info *pipe)
768{
769 smp_mb();
770 if (waitqueue_active(&pipe->wait))
771 wake_up_interruptible(&pipe->wait);
772 kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
773}
774
775/**
776 * splice_from_pipe_feed - feed available data from a pipe to a file
777 * @pipe: pipe to splice from
778 * @sd: information to @actor
779 * @actor: handler that splices the data
780 *
781 * Description:
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200782 * This function loops over the pipe and calls @actor to do the
783 * actual moving of a single struct pipe_buffer to the desired
784 * destination. It returns when there's no more buffers left in
785 * the pipe or if the requested number of bytes (@sd->total_len)
786 * have been copied. It returns a positive number (one) if the
787 * pipe needs to be filled with more data, zero if the required
788 * number of bytes have been copied and -errno on error.
789 *
790 * This, together with splice_from_pipe_{begin,end,next}, may be
791 * used to implement the functionality of __splice_from_pipe() when
792 * locking is required around copying the pipe buffers to the
793 * destination.
794 */
795int splice_from_pipe_feed(struct pipe_inode_info *pipe, struct splice_desc *sd,
796 splice_actor *actor)
797{
798 int ret;
799
800 while (pipe->nrbufs) {
801 struct pipe_buffer *buf = pipe->bufs + pipe->curbuf;
802 const struct pipe_buf_operations *ops = buf->ops;
803
804 sd->len = buf->len;
805 if (sd->len > sd->total_len)
806 sd->len = sd->total_len;
807
Michał Mirosława8adbe32010-12-17 08:56:44 +0100808 ret = buf->ops->confirm(pipe, buf);
809 if (unlikely(ret)) {
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200810 if (ret == -ENODATA)
811 ret = 0;
812 return ret;
813 }
Michał Mirosława8adbe32010-12-17 08:56:44 +0100814
815 ret = actor(pipe, buf, sd);
816 if (ret <= 0)
817 return ret;
818
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200819 buf->offset += ret;
820 buf->len -= ret;
821
822 sd->num_spliced += ret;
823 sd->len -= ret;
824 sd->pos += ret;
825 sd->total_len -= ret;
826
827 if (!buf->len) {
828 buf->ops = NULL;
829 ops->release(pipe, buf);
Jens Axboe35f3d142010-05-20 10:43:18 +0200830 pipe->curbuf = (pipe->curbuf + 1) & (pipe->buffers - 1);
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200831 pipe->nrbufs--;
Al Viro6447a3c2013-03-21 11:01:38 -0400832 if (pipe->files)
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200833 sd->need_wakeup = true;
834 }
835
836 if (!sd->total_len)
837 return 0;
838 }
839
840 return 1;
841}
842EXPORT_SYMBOL(splice_from_pipe_feed);
843
844/**
845 * splice_from_pipe_next - wait for some data to splice from
846 * @pipe: pipe to splice from
847 * @sd: information about the splice operation
848 *
849 * Description:
850 * This function will wait for some data and return a positive
851 * value (one) if pipe buffers are available. It will return zero
852 * or -errno if no more data needs to be spliced.
853 */
854int splice_from_pipe_next(struct pipe_inode_info *pipe, struct splice_desc *sd)
855{
856 while (!pipe->nrbufs) {
857 if (!pipe->writers)
858 return 0;
859
860 if (!pipe->waiting_writers && sd->num_spliced)
861 return 0;
862
863 if (sd->flags & SPLICE_F_NONBLOCK)
864 return -EAGAIN;
865
866 if (signal_pending(current))
867 return -ERESTARTSYS;
868
869 if (sd->need_wakeup) {
870 wakeup_pipe_writers(pipe);
871 sd->need_wakeup = false;
872 }
873
874 pipe_wait(pipe);
875 }
876
877 return 1;
878}
879EXPORT_SYMBOL(splice_from_pipe_next);
880
881/**
882 * splice_from_pipe_begin - start splicing from pipe
Randy Dunlapb80901b2009-04-16 19:09:55 -0700883 * @sd: information about the splice operation
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200884 *
885 * Description:
886 * This function should be called before a loop containing
887 * splice_from_pipe_next() and splice_from_pipe_feed() to
888 * initialize the necessary fields of @sd.
889 */
890void splice_from_pipe_begin(struct splice_desc *sd)
891{
892 sd->num_spliced = 0;
893 sd->need_wakeup = false;
894}
895EXPORT_SYMBOL(splice_from_pipe_begin);
896
897/**
898 * splice_from_pipe_end - finish splicing from pipe
899 * @pipe: pipe to splice from
900 * @sd: information about the splice operation
901 *
902 * Description:
903 * This function will wake up pipe writers if necessary. It should
904 * be called after a loop containing splice_from_pipe_next() and
905 * splice_from_pipe_feed().
906 */
907void splice_from_pipe_end(struct pipe_inode_info *pipe, struct splice_desc *sd)
908{
909 if (sd->need_wakeup)
910 wakeup_pipe_writers(pipe);
911}
912EXPORT_SYMBOL(splice_from_pipe_end);
913
Jens Axboe932cc6d2007-06-21 13:10:21 +0200914/**
915 * __splice_from_pipe - splice data from a pipe to given actor
916 * @pipe: pipe to splice from
917 * @sd: information to @actor
918 * @actor: handler that splices the data
919 *
920 * Description:
921 * This function does little more than loop over the pipe and call
922 * @actor to do the actual moving of a single struct pipe_buffer to
923 * the desired destination. See pipe_to_file, pipe_to_sendpage, or
924 * pipe_to_user.
925 *
Jens Axboe83f91352006-04-02 23:05:09 +0200926 */
Jens Axboec66ab6f2007-06-12 21:17:17 +0200927ssize_t __splice_from_pipe(struct pipe_inode_info *pipe, struct splice_desc *sd,
928 splice_actor *actor)
Jens Axboe5274f052006-03-30 15:15:30 +0200929{
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200930 int ret;
Jens Axboe5274f052006-03-30 15:15:30 +0200931
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200932 splice_from_pipe_begin(sd);
933 do {
934 ret = splice_from_pipe_next(pipe, sd);
935 if (ret > 0)
936 ret = splice_from_pipe_feed(pipe, sd, actor);
937 } while (ret > 0);
938 splice_from_pipe_end(pipe, sd);
Jens Axboe5274f052006-03-30 15:15:30 +0200939
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200940 return sd->num_spliced ? sd->num_spliced : ret;
Jens Axboe5274f052006-03-30 15:15:30 +0200941}
Mark Fasheh40bee44e2007-03-21 13:11:02 +0100942EXPORT_SYMBOL(__splice_from_pipe);
Jens Axboe5274f052006-03-30 15:15:30 +0200943
Jens Axboe932cc6d2007-06-21 13:10:21 +0200944/**
945 * splice_from_pipe - splice data from a pipe to a file
946 * @pipe: pipe to splice from
947 * @out: file to splice to
948 * @ppos: position in @out
949 * @len: how many bytes to splice
950 * @flags: splice modifier flags
951 * @actor: handler that splices the data
952 *
953 * Description:
Miklos Szeredi29339702009-04-14 19:48:37 +0200954 * See __splice_from_pipe. This function locks the pipe inode,
Jens Axboe932cc6d2007-06-21 13:10:21 +0200955 * otherwise it's identical to __splice_from_pipe().
956 *
957 */
Mark Fasheh6da61802006-10-17 18:43:07 +0200958ssize_t splice_from_pipe(struct pipe_inode_info *pipe, struct file *out,
959 loff_t *ppos, size_t len, unsigned int flags,
960 splice_actor *actor)
961{
962 ssize_t ret;
Jens Axboec66ab6f2007-06-12 21:17:17 +0200963 struct splice_desc sd = {
964 .total_len = len,
965 .flags = flags,
966 .pos = *ppos,
Jens Axboe6a14b902007-06-14 13:08:55 +0200967 .u.file = out,
Jens Axboec66ab6f2007-06-12 21:17:17 +0200968 };
Mark Fasheh6da61802006-10-17 18:43:07 +0200969
Miklos Szeredi61e0d472009-04-14 19:48:41 +0200970 pipe_lock(pipe);
Jens Axboec66ab6f2007-06-12 21:17:17 +0200971 ret = __splice_from_pipe(pipe, &sd, actor);
Miklos Szeredi61e0d472009-04-14 19:48:41 +0200972 pipe_unlock(pipe);
Mark Fasheh6da61802006-10-17 18:43:07 +0200973
974 return ret;
975}
976
977/**
Jens Axboe83f91352006-04-02 23:05:09 +0200978 * generic_file_splice_write - splice data from a pipe to a file
Ingo Molnar3a326a22006-04-10 15:18:35 +0200979 * @pipe: pipe info
Jens Axboe83f91352006-04-02 23:05:09 +0200980 * @out: file to write to
Jens Axboe932cc6d2007-06-21 13:10:21 +0200981 * @ppos: position in @out
Jens Axboe83f91352006-04-02 23:05:09 +0200982 * @len: number of bytes to splice
983 * @flags: splice modifier flags
984 *
Jens Axboe932cc6d2007-06-21 13:10:21 +0200985 * Description:
986 * Will either move or copy pages (determined by @flags options) from
987 * the given pipe inode to the given file.
Jens Axboe83f91352006-04-02 23:05:09 +0200988 *
989 */
Ingo Molnar3a326a22006-04-10 15:18:35 +0200990ssize_t
991generic_file_splice_write(struct pipe_inode_info *pipe, struct file *out,
Jens Axboecbb7e572006-04-11 14:57:50 +0200992 loff_t *ppos, size_t len, unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +0200993{
Jens Axboe4f6f0bd2006-04-02 23:04:46 +0200994 struct address_space *mapping = out->f_mapping;
Jens Axboe8c34e2d2006-10-17 19:43:22 +0200995 struct inode *inode = mapping->host;
Miklos Szeredi7f3d4ee2008-05-07 09:22:39 +0200996 struct splice_desc sd = {
997 .total_len = len,
998 .flags = flags,
999 .pos = *ppos,
1000 .u.file = out,
1001 };
Ingo Molnar3a326a22006-04-10 15:18:35 +02001002 ssize_t ret;
Jens Axboe8c34e2d2006-10-17 19:43:22 +02001003
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001004 pipe_lock(pipe);
Miklos Szeredieb443e52009-04-14 19:48:38 +02001005
1006 splice_from_pipe_begin(&sd);
1007 do {
1008 ret = splice_from_pipe_next(pipe, &sd);
1009 if (ret <= 0)
1010 break;
1011
1012 mutex_lock_nested(&inode->i_mutex, I_MUTEX_CHILD);
1013 ret = file_remove_suid(out);
Miklos Szeredi723590e2009-08-15 08:43:22 +02001014 if (!ret) {
Josef Bacikc3b2da32012-03-26 09:59:21 -04001015 ret = file_update_time(out);
1016 if (!ret)
1017 ret = splice_from_pipe_feed(pipe, &sd,
1018 pipe_to_file);
Miklos Szeredi723590e2009-08-15 08:43:22 +02001019 }
Miklos Szeredieb443e52009-04-14 19:48:38 +02001020 mutex_unlock(&inode->i_mutex);
1021 } while (ret > 0);
1022 splice_from_pipe_end(pipe, &sd);
1023
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001024 pipe_unlock(pipe);
Miklos Szeredieb443e52009-04-14 19:48:38 +02001025
1026 if (sd.num_spliced)
1027 ret = sd.num_spliced;
1028
Jens Axboea4514eb2006-04-19 15:57:05 +02001029 if (ret > 0) {
Jan Kara148f9482009-08-17 19:52:36 +02001030 int err;
Jens Axboe17ee4f42007-06-15 13:10:37 +02001031
Jan Kara148f9482009-08-17 19:52:36 +02001032 err = generic_write_sync(out, *ppos, ret);
1033 if (err)
1034 ret = err;
1035 else
1036 *ppos += ret;
Namjae Jeond0e1d662012-12-11 16:00:21 -08001037 balance_dirty_pages_ratelimited(mapping);
Jens Axboe4f6f0bd2006-04-02 23:04:46 +02001038 }
1039
1040 return ret;
Jens Axboe5274f052006-03-30 15:15:30 +02001041}
1042
Jens Axboe059a8f32006-04-02 23:06:05 +02001043EXPORT_SYMBOL(generic_file_splice_write);
1044
Miklos Szeredib2858d72009-05-19 11:37:46 +02001045static int write_pipe_buf(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
1046 struct splice_desc *sd)
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +02001047{
Miklos Szeredib2858d72009-05-19 11:37:46 +02001048 int ret;
1049 void *data;
Al Viro06ae43f2013-03-20 13:19:30 -04001050 loff_t tmp = sd->pos;
Miklos Szeredib2858d72009-05-19 11:37:46 +02001051
Miklos Szeredib2858d72009-05-19 11:37:46 +02001052 data = buf->ops->map(pipe, buf, 0);
Al Viro06ae43f2013-03-20 13:19:30 -04001053 ret = __kernel_write(sd->u.file, data + buf->offset, sd->len, &tmp);
Miklos Szeredib2858d72009-05-19 11:37:46 +02001054 buf->ops->unmap(pipe, buf, data);
1055
1056 return ret;
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +02001057}
1058
1059static ssize_t default_file_splice_write(struct pipe_inode_info *pipe,
1060 struct file *out, loff_t *ppos,
1061 size_t len, unsigned int flags)
1062{
Miklos Szeredib2858d72009-05-19 11:37:46 +02001063 ssize_t ret;
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +02001064
Miklos Szeredib2858d72009-05-19 11:37:46 +02001065 ret = splice_from_pipe(pipe, out, ppos, len, flags, write_pipe_buf);
1066 if (ret > 0)
1067 *ppos += ret;
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +02001068
Miklos Szeredib2858d72009-05-19 11:37:46 +02001069 return ret;
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +02001070}
1071
Jens Axboe83f91352006-04-02 23:05:09 +02001072/**
1073 * generic_splice_sendpage - splice data from a pipe to a socket
Jens Axboe932cc6d2007-06-21 13:10:21 +02001074 * @pipe: pipe to splice from
Jens Axboe83f91352006-04-02 23:05:09 +02001075 * @out: socket to write to
Jens Axboe932cc6d2007-06-21 13:10:21 +02001076 * @ppos: position in @out
Jens Axboe83f91352006-04-02 23:05:09 +02001077 * @len: number of bytes to splice
1078 * @flags: splice modifier flags
1079 *
Jens Axboe932cc6d2007-06-21 13:10:21 +02001080 * Description:
1081 * Will send @len bytes from the pipe to a network socket. No data copying
1082 * is involved.
Jens Axboe83f91352006-04-02 23:05:09 +02001083 *
1084 */
Ingo Molnar3a326a22006-04-10 15:18:35 +02001085ssize_t generic_splice_sendpage(struct pipe_inode_info *pipe, struct file *out,
Jens Axboecbb7e572006-04-11 14:57:50 +02001086 loff_t *ppos, size_t len, unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +02001087{
Jens Axboe00522fb2006-04-26 14:39:29 +02001088 return splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_sendpage);
Jens Axboe5274f052006-03-30 15:15:30 +02001089}
1090
Jens Axboe059a8f32006-04-02 23:06:05 +02001091EXPORT_SYMBOL(generic_splice_sendpage);
Jeff Garzika0f06782006-03-30 23:06:13 -05001092
Jens Axboe83f91352006-04-02 23:05:09 +02001093/*
1094 * Attempt to initiate a splice from pipe to file.
1095 */
Ingo Molnar3a326a22006-04-10 15:18:35 +02001096static long do_splice_from(struct pipe_inode_info *pipe, struct file *out,
Jens Axboecbb7e572006-04-11 14:57:50 +02001097 loff_t *ppos, size_t len, unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +02001098{
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +02001099 ssize_t (*splice_write)(struct pipe_inode_info *, struct file *,
1100 loff_t *, size_t, unsigned int);
Jens Axboe5274f052006-03-30 15:15:30 +02001101 int ret;
1102
Jens Axboe49570e92006-04-11 13:56:09 +02001103 if (unlikely(!(out->f_mode & FMODE_WRITE)))
Jens Axboe5274f052006-03-30 15:15:30 +02001104 return -EBADF;
1105
Linus Torvaldsefc968d2008-10-09 14:04:54 -07001106 if (unlikely(out->f_flags & O_APPEND))
1107 return -EINVAL;
1108
Jens Axboecbb7e572006-04-11 14:57:50 +02001109 ret = rw_verify_area(WRITE, out, ppos, len);
Jens Axboe5274f052006-03-30 15:15:30 +02001110 if (unlikely(ret < 0))
1111 return ret;
1112
Changli Gaocc56f7d2009-11-04 09:09:52 +01001113 if (out->f_op && out->f_op->splice_write)
1114 splice_write = out->f_op->splice_write;
1115 else
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +02001116 splice_write = default_file_splice_write;
1117
Al Viro2dd8c9a2013-03-20 13:21:32 -04001118 file_start_write(out);
1119 ret = splice_write(pipe, out, ppos, len, flags);
1120 file_end_write(out);
1121 return ret;
Jens Axboe5274f052006-03-30 15:15:30 +02001122}
1123
Jens Axboe83f91352006-04-02 23:05:09 +02001124/*
1125 * Attempt to initiate a splice from a file to a pipe.
1126 */
Jens Axboecbb7e572006-04-11 14:57:50 +02001127static long do_splice_to(struct file *in, loff_t *ppos,
1128 struct pipe_inode_info *pipe, size_t len,
1129 unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +02001130{
Miklos Szeredi68181732009-05-07 15:37:36 +02001131 ssize_t (*splice_read)(struct file *, loff_t *,
1132 struct pipe_inode_info *, size_t, unsigned int);
Jens Axboe5274f052006-03-30 15:15:30 +02001133 int ret;
1134
Jens Axboe49570e92006-04-11 13:56:09 +02001135 if (unlikely(!(in->f_mode & FMODE_READ)))
Jens Axboe5274f052006-03-30 15:15:30 +02001136 return -EBADF;
1137
Jens Axboecbb7e572006-04-11 14:57:50 +02001138 ret = rw_verify_area(READ, in, ppos, len);
Jens Axboe5274f052006-03-30 15:15:30 +02001139 if (unlikely(ret < 0))
1140 return ret;
1141
Changli Gaocc56f7d2009-11-04 09:09:52 +01001142 if (in->f_op && in->f_op->splice_read)
1143 splice_read = in->f_op->splice_read;
1144 else
Miklos Szeredi68181732009-05-07 15:37:36 +02001145 splice_read = default_file_splice_read;
1146
1147 return splice_read(in, ppos, pipe, len, flags);
Jens Axboe5274f052006-03-30 15:15:30 +02001148}
1149
Jens Axboe932cc6d2007-06-21 13:10:21 +02001150/**
1151 * splice_direct_to_actor - splices data directly between two non-pipes
1152 * @in: file to splice from
1153 * @sd: actor information on where to splice to
1154 * @actor: handles the data splicing
1155 *
1156 * Description:
1157 * This is a special case helper to splice directly between two
1158 * points, without requiring an explicit pipe. Internally an allocated
Randy Dunlap79685b82007-07-27 08:08:51 +02001159 * pipe is cached in the process, and reused during the lifetime of
Jens Axboe932cc6d2007-06-21 13:10:21 +02001160 * that process.
1161 *
Jens Axboec66ab6f2007-06-12 21:17:17 +02001162 */
1163ssize_t splice_direct_to_actor(struct file *in, struct splice_desc *sd,
1164 splice_direct_actor *actor)
Jens Axboeb92ce552006-04-11 13:52:07 +02001165{
1166 struct pipe_inode_info *pipe;
1167 long ret, bytes;
1168 umode_t i_mode;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001169 size_t len;
1170 int i, flags;
Jens Axboeb92ce552006-04-11 13:52:07 +02001171
1172 /*
1173 * We require the input being a regular file, as we don't want to
1174 * randomly drop data for eg socket -> socket splicing. Use the
1175 * piped splicing for that!
1176 */
Al Viro496ad9a2013-01-23 17:07:38 -05001177 i_mode = file_inode(in)->i_mode;
Jens Axboeb92ce552006-04-11 13:52:07 +02001178 if (unlikely(!S_ISREG(i_mode) && !S_ISBLK(i_mode)))
1179 return -EINVAL;
1180
1181 /*
1182 * neither in nor out is a pipe, setup an internal pipe attached to
1183 * 'out' and transfer the wanted data from 'in' to 'out' through that
1184 */
1185 pipe = current->splice_pipe;
Jens Axboe49570e92006-04-11 13:56:09 +02001186 if (unlikely(!pipe)) {
Al Viro7bee1302013-03-21 11:04:15 -04001187 pipe = alloc_pipe_info();
Jens Axboeb92ce552006-04-11 13:52:07 +02001188 if (!pipe)
1189 return -ENOMEM;
1190
1191 /*
1192 * We don't have an immediate reader, but we'll read the stuff
Jens Axboe00522fb2006-04-26 14:39:29 +02001193 * out of the pipe right after the splice_to_pipe(). So set
Jens Axboeb92ce552006-04-11 13:52:07 +02001194 * PIPE_READERS appropriately.
1195 */
1196 pipe->readers = 1;
1197
1198 current->splice_pipe = pipe;
1199 }
1200
1201 /*
Ingo Molnar73d62d82006-04-11 13:57:21 +02001202 * Do the splice.
Jens Axboeb92ce552006-04-11 13:52:07 +02001203 */
1204 ret = 0;
1205 bytes = 0;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001206 len = sd->total_len;
1207 flags = sd->flags;
1208
1209 /*
1210 * Don't block on output, we have to drain the direct pipe.
1211 */
1212 sd->flags &= ~SPLICE_F_NONBLOCK;
Jens Axboeb92ce552006-04-11 13:52:07 +02001213
1214 while (len) {
Jens Axboe51a92c02007-07-13 14:11:43 +02001215 size_t read_len;
Tom Zanussia82c53a2008-05-09 13:28:36 +02001216 loff_t pos = sd->pos, prev_pos = pos;
Jens Axboeb92ce552006-04-11 13:52:07 +02001217
Jens Axboebcd4f3a2007-07-16 14:41:49 +02001218 ret = do_splice_to(in, &pos, pipe, len, flags);
Jens Axboe51a92c02007-07-13 14:11:43 +02001219 if (unlikely(ret <= 0))
Jens Axboeb92ce552006-04-11 13:52:07 +02001220 goto out_release;
1221
1222 read_len = ret;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001223 sd->total_len = read_len;
Jens Axboeb92ce552006-04-11 13:52:07 +02001224
1225 /*
1226 * NOTE: nonblocking mode only applies to the input. We
1227 * must not do the output in nonblocking mode as then we
1228 * could get stuck data in the internal pipe:
1229 */
Jens Axboec66ab6f2007-06-12 21:17:17 +02001230 ret = actor(pipe, sd);
Tom Zanussia82c53a2008-05-09 13:28:36 +02001231 if (unlikely(ret <= 0)) {
1232 sd->pos = prev_pos;
Jens Axboeb92ce552006-04-11 13:52:07 +02001233 goto out_release;
Tom Zanussia82c53a2008-05-09 13:28:36 +02001234 }
Jens Axboeb92ce552006-04-11 13:52:07 +02001235
1236 bytes += ret;
1237 len -= ret;
Jens Axboebcd4f3a2007-07-16 14:41:49 +02001238 sd->pos = pos;
Jens Axboeb92ce552006-04-11 13:52:07 +02001239
Tom Zanussia82c53a2008-05-09 13:28:36 +02001240 if (ret < read_len) {
1241 sd->pos = prev_pos + ret;
Jens Axboe51a92c02007-07-13 14:11:43 +02001242 goto out_release;
Tom Zanussia82c53a2008-05-09 13:28:36 +02001243 }
Jens Axboeb92ce552006-04-11 13:52:07 +02001244 }
1245
Jens Axboe9e971982008-01-29 21:05:57 +01001246done:
Jens Axboeb92ce552006-04-11 13:52:07 +02001247 pipe->nrbufs = pipe->curbuf = 0;
Jens Axboe80848702008-01-30 12:24:48 +01001248 file_accessed(in);
Jens Axboeb92ce552006-04-11 13:52:07 +02001249 return bytes;
1250
1251out_release:
1252 /*
1253 * If we did an incomplete transfer we must release
1254 * the pipe buffers in question:
1255 */
Jens Axboe35f3d142010-05-20 10:43:18 +02001256 for (i = 0; i < pipe->buffers; i++) {
Jens Axboeb92ce552006-04-11 13:52:07 +02001257 struct pipe_buffer *buf = pipe->bufs + i;
1258
1259 if (buf->ops) {
1260 buf->ops->release(pipe, buf);
1261 buf->ops = NULL;
1262 }
1263 }
Jens Axboeb92ce552006-04-11 13:52:07 +02001264
Jens Axboe9e971982008-01-29 21:05:57 +01001265 if (!bytes)
1266 bytes = ret;
Jens Axboeb92ce552006-04-11 13:52:07 +02001267
Jens Axboe9e971982008-01-29 21:05:57 +01001268 goto done;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001269}
1270EXPORT_SYMBOL(splice_direct_to_actor);
1271
1272static int direct_splice_actor(struct pipe_inode_info *pipe,
1273 struct splice_desc *sd)
1274{
Jens Axboe6a14b902007-06-14 13:08:55 +02001275 struct file *file = sd->u.file;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001276
Changli Gao2cb4b052010-06-29 13:09:18 +02001277 return do_splice_from(pipe, file, &file->f_pos, sd->total_len,
1278 sd->flags);
Jens Axboec66ab6f2007-06-12 21:17:17 +02001279}
1280
Jens Axboe932cc6d2007-06-21 13:10:21 +02001281/**
1282 * do_splice_direct - splices data directly between two files
1283 * @in: file to splice from
1284 * @ppos: input file offset
1285 * @out: file to splice to
1286 * @len: number of bytes to splice
1287 * @flags: splice modifier flags
1288 *
1289 * Description:
1290 * For use by do_sendfile(). splice can easily emulate sendfile, but
1291 * doing it in the application would incur an extra system call
1292 * (splice in + splice out, as compared to just sendfile()). So this helper
1293 * can splice directly through a process-private pipe.
1294 *
1295 */
Jens Axboec66ab6f2007-06-12 21:17:17 +02001296long do_splice_direct(struct file *in, loff_t *ppos, struct file *out,
1297 size_t len, unsigned int flags)
1298{
1299 struct splice_desc sd = {
1300 .len = len,
1301 .total_len = len,
1302 .flags = flags,
1303 .pos = *ppos,
Jens Axboe6a14b902007-06-14 13:08:55 +02001304 .u.file = out,
Jens Axboec66ab6f2007-06-12 21:17:17 +02001305 };
Jens Axboe51a92c02007-07-13 14:11:43 +02001306 long ret;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001307
1308 ret = splice_direct_to_actor(in, &sd, direct_splice_actor);
Jens Axboe51a92c02007-07-13 14:11:43 +02001309 if (ret > 0)
Tom Zanussia82c53a2008-05-09 13:28:36 +02001310 *ppos = sd.pos;
Jens Axboe51a92c02007-07-13 14:11:43 +02001311
Jens Axboec66ab6f2007-06-12 21:17:17 +02001312 return ret;
Jens Axboeb92ce552006-04-11 13:52:07 +02001313}
1314
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001315static int splice_pipe_to_pipe(struct pipe_inode_info *ipipe,
1316 struct pipe_inode_info *opipe,
1317 size_t len, unsigned int flags);
Jens Axboeddac0d32006-11-04 12:49:32 +01001318
1319/*
Jens Axboe83f91352006-04-02 23:05:09 +02001320 * Determine where to splice to/from.
1321 */
Ingo Molnar529565d2006-04-10 15:18:58 +02001322static long do_splice(struct file *in, loff_t __user *off_in,
1323 struct file *out, loff_t __user *off_out,
1324 size_t len, unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +02001325{
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001326 struct pipe_inode_info *ipipe;
1327 struct pipe_inode_info *opipe;
Jens Axboecbb7e572006-04-11 14:57:50 +02001328 loff_t offset, *off;
Jens Axboea4514eb2006-04-19 15:57:05 +02001329 long ret;
Jens Axboe5274f052006-03-30 15:15:30 +02001330
Linus Torvalds71993e62010-11-28 13:56:09 -08001331 ipipe = get_pipe_info(in);
1332 opipe = get_pipe_info(out);
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001333
1334 if (ipipe && opipe) {
1335 if (off_in || off_out)
1336 return -ESPIPE;
1337
1338 if (!(in->f_mode & FMODE_READ))
1339 return -EBADF;
1340
1341 if (!(out->f_mode & FMODE_WRITE))
1342 return -EBADF;
1343
1344 /* Splicing to self would be fun, but... */
1345 if (ipipe == opipe)
1346 return -EINVAL;
1347
1348 return splice_pipe_to_pipe(ipipe, opipe, len, flags);
1349 }
1350
1351 if (ipipe) {
Ingo Molnar529565d2006-04-10 15:18:58 +02001352 if (off_in)
1353 return -ESPIPE;
Jens Axboeb92ce552006-04-11 13:52:07 +02001354 if (off_out) {
Changli Gao19c9a492010-06-29 13:10:36 +02001355 if (!(out->f_mode & FMODE_PWRITE))
Jens Axboeb92ce552006-04-11 13:52:07 +02001356 return -EINVAL;
Jens Axboecbb7e572006-04-11 14:57:50 +02001357 if (copy_from_user(&offset, off_out, sizeof(loff_t)))
Jens Axboeb92ce552006-04-11 13:52:07 +02001358 return -EFAULT;
Jens Axboecbb7e572006-04-11 14:57:50 +02001359 off = &offset;
1360 } else
1361 off = &out->f_pos;
Ingo Molnar529565d2006-04-10 15:18:58 +02001362
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001363 ret = do_splice_from(ipipe, out, off, len, flags);
Jens Axboea4514eb2006-04-19 15:57:05 +02001364
1365 if (off_out && copy_to_user(off_out, off, sizeof(loff_t)))
1366 ret = -EFAULT;
1367
1368 return ret;
Ingo Molnar529565d2006-04-10 15:18:58 +02001369 }
Jens Axboe5274f052006-03-30 15:15:30 +02001370
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001371 if (opipe) {
Ingo Molnar529565d2006-04-10 15:18:58 +02001372 if (off_out)
1373 return -ESPIPE;
Jens Axboeb92ce552006-04-11 13:52:07 +02001374 if (off_in) {
Changli Gao19c9a492010-06-29 13:10:36 +02001375 if (!(in->f_mode & FMODE_PREAD))
Jens Axboeb92ce552006-04-11 13:52:07 +02001376 return -EINVAL;
Jens Axboecbb7e572006-04-11 14:57:50 +02001377 if (copy_from_user(&offset, off_in, sizeof(loff_t)))
Jens Axboeb92ce552006-04-11 13:52:07 +02001378 return -EFAULT;
Jens Axboecbb7e572006-04-11 14:57:50 +02001379 off = &offset;
1380 } else
1381 off = &in->f_pos;
Ingo Molnar529565d2006-04-10 15:18:58 +02001382
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001383 ret = do_splice_to(in, off, opipe, len, flags);
Jens Axboea4514eb2006-04-19 15:57:05 +02001384
1385 if (off_in && copy_to_user(off_in, off, sizeof(loff_t)))
1386 ret = -EFAULT;
1387
1388 return ret;
Ingo Molnar529565d2006-04-10 15:18:58 +02001389 }
Jens Axboe5274f052006-03-30 15:15:30 +02001390
1391 return -EINVAL;
1392}
1393
Jens Axboe912d35f2006-04-26 10:59:21 +02001394/*
1395 * Map an iov into an array of pages and offset/length tupples. With the
1396 * partial_page structure, we can map several non-contiguous ranges into
1397 * our ones pages[] map instead of splitting that operation into pieces.
1398 * Could easily be exported as a generic helper for other users, in which
1399 * case one would probably want to add a 'max_nr_pages' parameter as well.
1400 */
1401static int get_iovec_page_array(const struct iovec __user *iov,
1402 unsigned int nr_vecs, struct page **pages,
Eric Dumazetbd1a68b2012-04-04 11:08:51 +02001403 struct partial_page *partial, bool aligned,
Jens Axboe35f3d142010-05-20 10:43:18 +02001404 unsigned int pipe_buffers)
Jens Axboe912d35f2006-04-26 10:59:21 +02001405{
1406 int buffers = 0, error = 0;
1407
Jens Axboe912d35f2006-04-26 10:59:21 +02001408 while (nr_vecs) {
1409 unsigned long off, npages;
Linus Torvalds75723952007-10-01 13:17:28 -07001410 struct iovec entry;
Jens Axboe912d35f2006-04-26 10:59:21 +02001411 void __user *base;
1412 size_t len;
1413 int i;
1414
Linus Torvalds75723952007-10-01 13:17:28 -07001415 error = -EFAULT;
Nick Pigginbc40d732008-07-25 19:45:26 -07001416 if (copy_from_user(&entry, iov, sizeof(entry)))
Jens Axboe912d35f2006-04-26 10:59:21 +02001417 break;
Linus Torvalds75723952007-10-01 13:17:28 -07001418
1419 base = entry.iov_base;
1420 len = entry.iov_len;
Jens Axboe912d35f2006-04-26 10:59:21 +02001421
1422 /*
1423 * Sanity check this iovec. 0 read succeeds.
1424 */
Linus Torvalds75723952007-10-01 13:17:28 -07001425 error = 0;
Jens Axboe912d35f2006-04-26 10:59:21 +02001426 if (unlikely(!len))
1427 break;
1428 error = -EFAULT;
Bastian Blank712a30e2008-02-10 16:47:57 +02001429 if (!access_ok(VERIFY_READ, base, len))
Jens Axboe912d35f2006-04-26 10:59:21 +02001430 break;
1431
1432 /*
1433 * Get this base offset and number of pages, then map
1434 * in the user pages.
1435 */
1436 off = (unsigned long) base & ~PAGE_MASK;
Jens Axboe7afa6fd2006-05-01 20:02:33 +02001437
1438 /*
1439 * If asked for alignment, the offset must be zero and the
1440 * length a multiple of the PAGE_SIZE.
1441 */
1442 error = -EINVAL;
1443 if (aligned && (off || len & ~PAGE_MASK))
1444 break;
1445
Jens Axboe912d35f2006-04-26 10:59:21 +02001446 npages = (off + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
Jens Axboe35f3d142010-05-20 10:43:18 +02001447 if (npages > pipe_buffers - buffers)
1448 npages = pipe_buffers - buffers;
Jens Axboe912d35f2006-04-26 10:59:21 +02001449
Nick Pigginbc40d732008-07-25 19:45:26 -07001450 error = get_user_pages_fast((unsigned long)base, npages,
1451 0, &pages[buffers]);
Jens Axboe912d35f2006-04-26 10:59:21 +02001452
1453 if (unlikely(error <= 0))
1454 break;
1455
1456 /*
1457 * Fill this contiguous range into the partial page map.
1458 */
1459 for (i = 0; i < error; i++) {
Jens Axboe75914892006-05-02 12:57:18 +02001460 const int plen = min_t(size_t, len, PAGE_SIZE - off);
Jens Axboe912d35f2006-04-26 10:59:21 +02001461
1462 partial[buffers].offset = off;
1463 partial[buffers].len = plen;
1464
1465 off = 0;
1466 len -= plen;
1467 buffers++;
1468 }
1469
1470 /*
1471 * We didn't complete this iov, stop here since it probably
1472 * means we have to move some of this into a pipe to
1473 * be able to continue.
1474 */
1475 if (len)
1476 break;
1477
1478 /*
1479 * Don't continue if we mapped fewer pages than we asked for,
1480 * or if we mapped the max number of pages that we have
1481 * room for.
1482 */
Jens Axboe35f3d142010-05-20 10:43:18 +02001483 if (error < npages || buffers == pipe_buffers)
Jens Axboe912d35f2006-04-26 10:59:21 +02001484 break;
1485
1486 nr_vecs--;
1487 iov++;
1488 }
1489
Jens Axboe912d35f2006-04-26 10:59:21 +02001490 if (buffers)
1491 return buffers;
1492
1493 return error;
1494}
1495
Jens Axboe6a14b902007-06-14 13:08:55 +02001496static int pipe_to_user(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
1497 struct splice_desc *sd)
1498{
1499 char *src;
1500 int ret;
1501
Jens Axboe6a14b902007-06-14 13:08:55 +02001502 /*
1503 * See if we can use the atomic maps, by prefaulting in the
1504 * pages and doing an atomic copy
1505 */
1506 if (!fault_in_pages_writeable(sd->u.userptr, sd->len)) {
1507 src = buf->ops->map(pipe, buf, 1);
1508 ret = __copy_to_user_inatomic(sd->u.userptr, src + buf->offset,
1509 sd->len);
1510 buf->ops->unmap(pipe, buf, src);
1511 if (!ret) {
1512 ret = sd->len;
1513 goto out;
1514 }
1515 }
1516
1517 /*
1518 * No dice, use slow non-atomic map and copy
1519 */
1520 src = buf->ops->map(pipe, buf, 0);
1521
1522 ret = sd->len;
1523 if (copy_to_user(sd->u.userptr, src + buf->offset, sd->len))
1524 ret = -EFAULT;
1525
Jens Axboe6866bef2007-10-16 10:01:29 +02001526 buf->ops->unmap(pipe, buf, src);
Jens Axboe6a14b902007-06-14 13:08:55 +02001527out:
1528 if (ret > 0)
1529 sd->u.userptr += ret;
Jens Axboe6a14b902007-06-14 13:08:55 +02001530 return ret;
1531}
1532
1533/*
1534 * For lack of a better implementation, implement vmsplice() to userspace
1535 * as a simple copy of the pipes pages to the user iov.
1536 */
1537static long vmsplice_to_user(struct file *file, const struct iovec __user *iov,
1538 unsigned long nr_segs, unsigned int flags)
1539{
1540 struct pipe_inode_info *pipe;
1541 struct splice_desc sd;
1542 ssize_t size;
1543 int error;
1544 long ret;
1545
Linus Torvalds71993e62010-11-28 13:56:09 -08001546 pipe = get_pipe_info(file);
Jens Axboe6a14b902007-06-14 13:08:55 +02001547 if (!pipe)
1548 return -EBADF;
1549
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001550 pipe_lock(pipe);
Jens Axboe6a14b902007-06-14 13:08:55 +02001551
1552 error = ret = 0;
1553 while (nr_segs) {
1554 void __user *base;
1555 size_t len;
1556
1557 /*
1558 * Get user address base and length for this iovec.
1559 */
1560 error = get_user(base, &iov->iov_base);
1561 if (unlikely(error))
1562 break;
1563 error = get_user(len, &iov->iov_len);
1564 if (unlikely(error))
1565 break;
1566
1567 /*
1568 * Sanity check this iovec. 0 read succeeds.
1569 */
1570 if (unlikely(!len))
1571 break;
1572 if (unlikely(!base)) {
1573 error = -EFAULT;
1574 break;
1575 }
1576
Jens Axboe88119302008-02-08 08:49:14 -08001577 if (unlikely(!access_ok(VERIFY_WRITE, base, len))) {
1578 error = -EFAULT;
1579 break;
1580 }
1581
Jens Axboe6a14b902007-06-14 13:08:55 +02001582 sd.len = 0;
1583 sd.total_len = len;
1584 sd.flags = flags;
1585 sd.u.userptr = base;
1586 sd.pos = 0;
1587
1588 size = __splice_from_pipe(pipe, &sd, pipe_to_user);
1589 if (size < 0) {
1590 if (!ret)
1591 ret = size;
1592
1593 break;
1594 }
1595
1596 ret += size;
1597
1598 if (size < len)
1599 break;
1600
1601 nr_segs--;
1602 iov++;
1603 }
1604
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001605 pipe_unlock(pipe);
Jens Axboe6a14b902007-06-14 13:08:55 +02001606
1607 if (!ret)
1608 ret = error;
1609
1610 return ret;
1611}
1612
Jens Axboe912d35f2006-04-26 10:59:21 +02001613/*
1614 * vmsplice splices a user address range into a pipe. It can be thought of
1615 * as splice-from-memory, where the regular splice is splice-from-file (or
1616 * to file). In both cases the output is a pipe, naturally.
Jens Axboe912d35f2006-04-26 10:59:21 +02001617 */
Jens Axboe6a14b902007-06-14 13:08:55 +02001618static long vmsplice_to_pipe(struct file *file, const struct iovec __user *iov,
1619 unsigned long nr_segs, unsigned int flags)
Jens Axboe912d35f2006-04-26 10:59:21 +02001620{
Jens Axboeddac0d32006-11-04 12:49:32 +01001621 struct pipe_inode_info *pipe;
Jens Axboe35f3d142010-05-20 10:43:18 +02001622 struct page *pages[PIPE_DEF_BUFFERS];
1623 struct partial_page partial[PIPE_DEF_BUFFERS];
Jens Axboe912d35f2006-04-26 10:59:21 +02001624 struct splice_pipe_desc spd = {
1625 .pages = pages,
1626 .partial = partial,
Eric Dumazet047fe362012-06-12 15:24:40 +02001627 .nr_pages_max = PIPE_DEF_BUFFERS,
Jens Axboe912d35f2006-04-26 10:59:21 +02001628 .flags = flags,
1629 .ops = &user_page_pipe_buf_ops,
Jens Axboebbdfc2f2007-11-06 23:29:47 -08001630 .spd_release = spd_release_page,
Jens Axboe912d35f2006-04-26 10:59:21 +02001631 };
Jens Axboe35f3d142010-05-20 10:43:18 +02001632 long ret;
Jens Axboe912d35f2006-04-26 10:59:21 +02001633
Linus Torvalds71993e62010-11-28 13:56:09 -08001634 pipe = get_pipe_info(file);
Jens Axboeddac0d32006-11-04 12:49:32 +01001635 if (!pipe)
Jens Axboe912d35f2006-04-26 10:59:21 +02001636 return -EBADF;
Jens Axboe912d35f2006-04-26 10:59:21 +02001637
Jens Axboe35f3d142010-05-20 10:43:18 +02001638 if (splice_grow_spd(pipe, &spd))
1639 return -ENOMEM;
Jens Axboe912d35f2006-04-26 10:59:21 +02001640
Jens Axboe35f3d142010-05-20 10:43:18 +02001641 spd.nr_pages = get_iovec_page_array(iov, nr_segs, spd.pages,
Eric Dumazetbd1a68b2012-04-04 11:08:51 +02001642 spd.partial, false,
Eric Dumazet047fe362012-06-12 15:24:40 +02001643 spd.nr_pages_max);
Jens Axboe35f3d142010-05-20 10:43:18 +02001644 if (spd.nr_pages <= 0)
1645 ret = spd.nr_pages;
1646 else
1647 ret = splice_to_pipe(pipe, &spd);
1648
Eric Dumazet047fe362012-06-12 15:24:40 +02001649 splice_shrink_spd(&spd);
Jens Axboe35f3d142010-05-20 10:43:18 +02001650 return ret;
Jens Axboe912d35f2006-04-26 10:59:21 +02001651}
1652
Jens Axboe6a14b902007-06-14 13:08:55 +02001653/*
1654 * Note that vmsplice only really supports true splicing _from_ user memory
1655 * to a pipe, not the other way around. Splicing from user memory is a simple
1656 * operation that can be supported without any funky alignment restrictions
1657 * or nasty vm tricks. We simply map in the user memory and fill them into
1658 * a pipe. The reverse isn't quite as easy, though. There are two possible
1659 * solutions for that:
1660 *
1661 * - memcpy() the data internally, at which point we might as well just
1662 * do a regular read() on the buffer anyway.
1663 * - Lots of nasty vm tricks, that are neither fast nor flexible (it
1664 * has restriction limitations on both ends of the pipe).
1665 *
1666 * Currently we punt and implement it as a normal copy, see pipe_to_user().
1667 *
1668 */
Heiko Carstens836f92a2009-01-14 14:14:33 +01001669SYSCALL_DEFINE4(vmsplice, int, fd, const struct iovec __user *, iov,
1670 unsigned long, nr_segs, unsigned int, flags)
Jens Axboe912d35f2006-04-26 10:59:21 +02001671{
Al Viro2903ff02012-08-28 12:52:22 -04001672 struct fd f;
Jens Axboe912d35f2006-04-26 10:59:21 +02001673 long error;
Jens Axboe912d35f2006-04-26 10:59:21 +02001674
Jens Axboe6a14b902007-06-14 13:08:55 +02001675 if (unlikely(nr_segs > UIO_MAXIOV))
1676 return -EINVAL;
1677 else if (unlikely(!nr_segs))
1678 return 0;
1679
Jens Axboe912d35f2006-04-26 10:59:21 +02001680 error = -EBADF;
Al Viro2903ff02012-08-28 12:52:22 -04001681 f = fdget(fd);
1682 if (f.file) {
1683 if (f.file->f_mode & FMODE_WRITE)
1684 error = vmsplice_to_pipe(f.file, iov, nr_segs, flags);
1685 else if (f.file->f_mode & FMODE_READ)
1686 error = vmsplice_to_user(f.file, iov, nr_segs, flags);
Jens Axboe912d35f2006-04-26 10:59:21 +02001687
Al Viro2903ff02012-08-28 12:52:22 -04001688 fdput(f);
Jens Axboe912d35f2006-04-26 10:59:21 +02001689 }
1690
1691 return error;
1692}
1693
Al Viro76b021d2013-03-02 10:19:56 -05001694#ifdef CONFIG_COMPAT
1695COMPAT_SYSCALL_DEFINE4(vmsplice, int, fd, const struct compat_iovec __user *, iov32,
1696 unsigned int, nr_segs, unsigned int, flags)
1697{
1698 unsigned i;
1699 struct iovec __user *iov;
1700 if (nr_segs > UIO_MAXIOV)
1701 return -EINVAL;
1702 iov = compat_alloc_user_space(nr_segs * sizeof(struct iovec));
1703 for (i = 0; i < nr_segs; i++) {
1704 struct compat_iovec v;
1705 if (get_user(v.iov_base, &iov32[i].iov_base) ||
1706 get_user(v.iov_len, &iov32[i].iov_len) ||
1707 put_user(compat_ptr(v.iov_base), &iov[i].iov_base) ||
1708 put_user(v.iov_len, &iov[i].iov_len))
1709 return -EFAULT;
1710 }
1711 return sys_vmsplice(fd, iov, nr_segs, flags);
1712}
1713#endif
1714
Heiko Carstens836f92a2009-01-14 14:14:33 +01001715SYSCALL_DEFINE6(splice, int, fd_in, loff_t __user *, off_in,
1716 int, fd_out, loff_t __user *, off_out,
1717 size_t, len, unsigned int, flags)
Jens Axboe5274f052006-03-30 15:15:30 +02001718{
Al Viro2903ff02012-08-28 12:52:22 -04001719 struct fd in, out;
Jens Axboe5274f052006-03-30 15:15:30 +02001720 long error;
Jens Axboe5274f052006-03-30 15:15:30 +02001721
1722 if (unlikely(!len))
1723 return 0;
1724
1725 error = -EBADF;
Al Viro2903ff02012-08-28 12:52:22 -04001726 in = fdget(fd_in);
1727 if (in.file) {
1728 if (in.file->f_mode & FMODE_READ) {
1729 out = fdget(fd_out);
1730 if (out.file) {
1731 if (out.file->f_mode & FMODE_WRITE)
1732 error = do_splice(in.file, off_in,
1733 out.file, off_out,
Ingo Molnar529565d2006-04-10 15:18:58 +02001734 len, flags);
Al Viro2903ff02012-08-28 12:52:22 -04001735 fdput(out);
Jens Axboe5274f052006-03-30 15:15:30 +02001736 }
1737 }
Al Viro2903ff02012-08-28 12:52:22 -04001738 fdput(in);
Jens Axboe5274f052006-03-30 15:15:30 +02001739 }
Jens Axboe5274f052006-03-30 15:15:30 +02001740 return error;
1741}
Jens Axboe70524492006-04-11 15:51:17 +02001742
1743/*
Jens Axboeaadd06e2006-07-10 11:00:01 +02001744 * Make sure there's data to read. Wait for input if we can, otherwise
1745 * return an appropriate error.
1746 */
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001747static int ipipe_prep(struct pipe_inode_info *pipe, unsigned int flags)
Jens Axboeaadd06e2006-07-10 11:00:01 +02001748{
1749 int ret;
1750
1751 /*
1752 * Check ->nrbufs without the inode lock first. This function
1753 * is speculative anyways, so missing one is ok.
1754 */
1755 if (pipe->nrbufs)
1756 return 0;
1757
1758 ret = 0;
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001759 pipe_lock(pipe);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001760
1761 while (!pipe->nrbufs) {
1762 if (signal_pending(current)) {
1763 ret = -ERESTARTSYS;
1764 break;
1765 }
1766 if (!pipe->writers)
1767 break;
1768 if (!pipe->waiting_writers) {
1769 if (flags & SPLICE_F_NONBLOCK) {
1770 ret = -EAGAIN;
1771 break;
1772 }
1773 }
1774 pipe_wait(pipe);
1775 }
1776
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001777 pipe_unlock(pipe);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001778 return ret;
1779}
1780
1781/*
1782 * Make sure there's writeable room. Wait for room if we can, otherwise
1783 * return an appropriate error.
1784 */
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001785static int opipe_prep(struct pipe_inode_info *pipe, unsigned int flags)
Jens Axboeaadd06e2006-07-10 11:00:01 +02001786{
1787 int ret;
1788
1789 /*
1790 * Check ->nrbufs without the inode lock first. This function
1791 * is speculative anyways, so missing one is ok.
1792 */
Jens Axboe35f3d142010-05-20 10:43:18 +02001793 if (pipe->nrbufs < pipe->buffers)
Jens Axboeaadd06e2006-07-10 11:00:01 +02001794 return 0;
1795
1796 ret = 0;
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001797 pipe_lock(pipe);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001798
Jens Axboe35f3d142010-05-20 10:43:18 +02001799 while (pipe->nrbufs >= pipe->buffers) {
Jens Axboeaadd06e2006-07-10 11:00:01 +02001800 if (!pipe->readers) {
1801 send_sig(SIGPIPE, current, 0);
1802 ret = -EPIPE;
1803 break;
1804 }
1805 if (flags & SPLICE_F_NONBLOCK) {
1806 ret = -EAGAIN;
1807 break;
1808 }
1809 if (signal_pending(current)) {
1810 ret = -ERESTARTSYS;
1811 break;
1812 }
1813 pipe->waiting_writers++;
1814 pipe_wait(pipe);
1815 pipe->waiting_writers--;
1816 }
1817
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001818 pipe_unlock(pipe);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001819 return ret;
1820}
1821
1822/*
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001823 * Splice contents of ipipe to opipe.
1824 */
1825static int splice_pipe_to_pipe(struct pipe_inode_info *ipipe,
1826 struct pipe_inode_info *opipe,
1827 size_t len, unsigned int flags)
1828{
1829 struct pipe_buffer *ibuf, *obuf;
1830 int ret = 0, nbuf;
1831 bool input_wakeup = false;
1832
1833
1834retry:
1835 ret = ipipe_prep(ipipe, flags);
1836 if (ret)
1837 return ret;
1838
1839 ret = opipe_prep(opipe, flags);
1840 if (ret)
1841 return ret;
1842
1843 /*
1844 * Potential ABBA deadlock, work around it by ordering lock
1845 * grabbing by pipe info address. Otherwise two different processes
1846 * could deadlock (one doing tee from A -> B, the other from B -> A).
1847 */
1848 pipe_double_lock(ipipe, opipe);
1849
1850 do {
1851 if (!opipe->readers) {
1852 send_sig(SIGPIPE, current, 0);
1853 if (!ret)
1854 ret = -EPIPE;
1855 break;
1856 }
1857
1858 if (!ipipe->nrbufs && !ipipe->writers)
1859 break;
1860
1861 /*
1862 * Cannot make any progress, because either the input
1863 * pipe is empty or the output pipe is full.
1864 */
Jens Axboe35f3d142010-05-20 10:43:18 +02001865 if (!ipipe->nrbufs || opipe->nrbufs >= opipe->buffers) {
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001866 /* Already processed some buffers, break */
1867 if (ret)
1868 break;
1869
1870 if (flags & SPLICE_F_NONBLOCK) {
1871 ret = -EAGAIN;
1872 break;
1873 }
1874
1875 /*
1876 * We raced with another reader/writer and haven't
1877 * managed to process any buffers. A zero return
1878 * value means EOF, so retry instead.
1879 */
1880 pipe_unlock(ipipe);
1881 pipe_unlock(opipe);
1882 goto retry;
1883 }
1884
1885 ibuf = ipipe->bufs + ipipe->curbuf;
Jens Axboe35f3d142010-05-20 10:43:18 +02001886 nbuf = (opipe->curbuf + opipe->nrbufs) & (opipe->buffers - 1);
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001887 obuf = opipe->bufs + nbuf;
1888
1889 if (len >= ibuf->len) {
1890 /*
1891 * Simply move the whole buffer from ipipe to opipe
1892 */
1893 *obuf = *ibuf;
1894 ibuf->ops = NULL;
1895 opipe->nrbufs++;
Jens Axboe35f3d142010-05-20 10:43:18 +02001896 ipipe->curbuf = (ipipe->curbuf + 1) & (ipipe->buffers - 1);
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001897 ipipe->nrbufs--;
1898 input_wakeup = true;
1899 } else {
1900 /*
1901 * Get a reference to this pipe buffer,
1902 * so we can copy the contents over.
1903 */
1904 ibuf->ops->get(ipipe, ibuf);
1905 *obuf = *ibuf;
1906
1907 /*
1908 * Don't inherit the gift flag, we need to
1909 * prevent multiple steals of this page.
1910 */
1911 obuf->flags &= ~PIPE_BUF_FLAG_GIFT;
1912
1913 obuf->len = len;
1914 opipe->nrbufs++;
1915 ibuf->offset += obuf->len;
1916 ibuf->len -= obuf->len;
1917 }
1918 ret += obuf->len;
1919 len -= obuf->len;
1920 } while (len);
1921
1922 pipe_unlock(ipipe);
1923 pipe_unlock(opipe);
1924
1925 /*
1926 * If we put data in the output pipe, wakeup any potential readers.
1927 */
Namhyung Kim825cdcb2011-05-23 19:58:53 +02001928 if (ret > 0)
1929 wakeup_pipe_readers(opipe);
1930
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001931 if (input_wakeup)
1932 wakeup_pipe_writers(ipipe);
1933
1934 return ret;
1935}
1936
1937/*
Jens Axboe70524492006-04-11 15:51:17 +02001938 * Link contents of ipipe to opipe.
1939 */
1940static int link_pipe(struct pipe_inode_info *ipipe,
1941 struct pipe_inode_info *opipe,
1942 size_t len, unsigned int flags)
1943{
1944 struct pipe_buffer *ibuf, *obuf;
Jens Axboeaadd06e2006-07-10 11:00:01 +02001945 int ret = 0, i = 0, nbuf;
Jens Axboe70524492006-04-11 15:51:17 +02001946
1947 /*
1948 * Potential ABBA deadlock, work around it by ordering lock
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001949 * grabbing by pipe info address. Otherwise two different processes
Jens Axboe70524492006-04-11 15:51:17 +02001950 * could deadlock (one doing tee from A -> B, the other from B -> A).
1951 */
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001952 pipe_double_lock(ipipe, opipe);
Jens Axboe70524492006-04-11 15:51:17 +02001953
Jens Axboeaadd06e2006-07-10 11:00:01 +02001954 do {
Jens Axboe70524492006-04-11 15:51:17 +02001955 if (!opipe->readers) {
1956 send_sig(SIGPIPE, current, 0);
1957 if (!ret)
1958 ret = -EPIPE;
1959 break;
1960 }
Jens Axboe70524492006-04-11 15:51:17 +02001961
1962 /*
Jens Axboeaadd06e2006-07-10 11:00:01 +02001963 * If we have iterated all input buffers or ran out of
1964 * output room, break.
Jens Axboe70524492006-04-11 15:51:17 +02001965 */
Jens Axboe35f3d142010-05-20 10:43:18 +02001966 if (i >= ipipe->nrbufs || opipe->nrbufs >= opipe->buffers)
Jens Axboe70524492006-04-11 15:51:17 +02001967 break;
Jens Axboeaadd06e2006-07-10 11:00:01 +02001968
Jens Axboe35f3d142010-05-20 10:43:18 +02001969 ibuf = ipipe->bufs + ((ipipe->curbuf + i) & (ipipe->buffers-1));
1970 nbuf = (opipe->curbuf + opipe->nrbufs) & (opipe->buffers - 1);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001971
Jens Axboe2a27250e2006-04-19 15:56:40 +02001972 /*
Jens Axboeaadd06e2006-07-10 11:00:01 +02001973 * Get a reference to this pipe buffer,
1974 * so we can copy the contents over.
Jens Axboe2a27250e2006-04-19 15:56:40 +02001975 */
Jens Axboeaadd06e2006-07-10 11:00:01 +02001976 ibuf->ops->get(ipipe, ibuf);
Jens Axboe70524492006-04-11 15:51:17 +02001977
Jens Axboeaadd06e2006-07-10 11:00:01 +02001978 obuf = opipe->bufs + nbuf;
1979 *obuf = *ibuf;
Jens Axboe70524492006-04-11 15:51:17 +02001980
Jens Axboeaadd06e2006-07-10 11:00:01 +02001981 /*
1982 * Don't inherit the gift flag, we need to
1983 * prevent multiple steals of this page.
1984 */
1985 obuf->flags &= ~PIPE_BUF_FLAG_GIFT;
1986
1987 if (obuf->len > len)
1988 obuf->len = len;
1989
1990 opipe->nrbufs++;
1991 ret += obuf->len;
1992 len -= obuf->len;
1993 i++;
1994 } while (len);
Jens Axboe70524492006-04-11 15:51:17 +02001995
Jens Axboe02cf01a2008-02-20 10:34:51 +01001996 /*
1997 * return EAGAIN if we have the potential of some data in the
1998 * future, otherwise just return 0
1999 */
2000 if (!ret && ipipe->waiting_writers && (flags & SPLICE_F_NONBLOCK))
2001 ret = -EAGAIN;
2002
Miklos Szeredi61e0d472009-04-14 19:48:41 +02002003 pipe_unlock(ipipe);
2004 pipe_unlock(opipe);
Jens Axboe70524492006-04-11 15:51:17 +02002005
Jens Axboeaadd06e2006-07-10 11:00:01 +02002006 /*
2007 * If we put data in the output pipe, wakeup any potential readers.
2008 */
Namhyung Kim825cdcb2011-05-23 19:58:53 +02002009 if (ret > 0)
2010 wakeup_pipe_readers(opipe);
Jens Axboe70524492006-04-11 15:51:17 +02002011
2012 return ret;
2013}
2014
2015/*
2016 * This is a tee(1) implementation that works on pipes. It doesn't copy
2017 * any data, it simply references the 'in' pages on the 'out' pipe.
2018 * The 'flags' used are the SPLICE_F_* variants, currently the only
2019 * applicable one is SPLICE_F_NONBLOCK.
2020 */
2021static long do_tee(struct file *in, struct file *out, size_t len,
2022 unsigned int flags)
2023{
Linus Torvalds71993e62010-11-28 13:56:09 -08002024 struct pipe_inode_info *ipipe = get_pipe_info(in);
2025 struct pipe_inode_info *opipe = get_pipe_info(out);
Jens Axboeaadd06e2006-07-10 11:00:01 +02002026 int ret = -EINVAL;
Jens Axboe70524492006-04-11 15:51:17 +02002027
2028 /*
Jens Axboeaadd06e2006-07-10 11:00:01 +02002029 * Duplicate the contents of ipipe to opipe without actually
2030 * copying the data.
Jens Axboe70524492006-04-11 15:51:17 +02002031 */
Jens Axboeaadd06e2006-07-10 11:00:01 +02002032 if (ipipe && opipe && ipipe != opipe) {
2033 /*
2034 * Keep going, unless we encounter an error. The ipipe/opipe
2035 * ordering doesn't really matter.
2036 */
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02002037 ret = ipipe_prep(ipipe, flags);
Jens Axboeaadd06e2006-07-10 11:00:01 +02002038 if (!ret) {
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02002039 ret = opipe_prep(opipe, flags);
Jens Axboe02cf01a2008-02-20 10:34:51 +01002040 if (!ret)
Jens Axboeaadd06e2006-07-10 11:00:01 +02002041 ret = link_pipe(ipipe, opipe, len, flags);
Jens Axboeaadd06e2006-07-10 11:00:01 +02002042 }
2043 }
Jens Axboe70524492006-04-11 15:51:17 +02002044
Jens Axboeaadd06e2006-07-10 11:00:01 +02002045 return ret;
Jens Axboe70524492006-04-11 15:51:17 +02002046}
2047
Heiko Carstens836f92a2009-01-14 14:14:33 +01002048SYSCALL_DEFINE4(tee, int, fdin, int, fdout, size_t, len, unsigned int, flags)
Jens Axboe70524492006-04-11 15:51:17 +02002049{
Al Viro2903ff02012-08-28 12:52:22 -04002050 struct fd in;
2051 int error;
Jens Axboe70524492006-04-11 15:51:17 +02002052
2053 if (unlikely(!len))
2054 return 0;
2055
2056 error = -EBADF;
Al Viro2903ff02012-08-28 12:52:22 -04002057 in = fdget(fdin);
2058 if (in.file) {
2059 if (in.file->f_mode & FMODE_READ) {
2060 struct fd out = fdget(fdout);
2061 if (out.file) {
2062 if (out.file->f_mode & FMODE_WRITE)
2063 error = do_tee(in.file, out.file,
2064 len, flags);
2065 fdput(out);
Jens Axboe70524492006-04-11 15:51:17 +02002066 }
2067 }
Al Viro2903ff02012-08-28 12:52:22 -04002068 fdput(in);
Jens Axboe70524492006-04-11 15:51:17 +02002069 }
2070
2071 return error;
2072}