blob: 6ec6054df138b0bd3804882ea40751d74673cbc1 [file] [log] [blame]
Jens Axboe5274f052006-03-30 15:15:30 +02001/*
2 * "splice": joining two ropes together by interweaving their strands.
3 *
4 * This is the "extended pipe" functionality, where a pipe is used as
5 * an arbitrary in-memory buffer. Think of a pipe as a small kernel
6 * buffer that you can use to transfer data from one end to the other.
7 *
8 * The traditional unix read/write is extended with a "splice()" operation
9 * that transfers data buffers to or from a pipe buffer.
10 *
11 * Named by Larry McVoy, original implementation from Linus, extended by
Jens Axboec2058e02006-04-11 13:56:34 +020012 * Jens to support splicing to files, network, direct splicing, etc and
13 * fixing lots of bugs.
Jens Axboe5274f052006-03-30 15:15:30 +020014 *
Jens Axboe0fe23472006-09-04 15:41:16 +020015 * Copyright (C) 2005-2006 Jens Axboe <axboe@kernel.dk>
Jens Axboec2058e02006-04-11 13:56:34 +020016 * Copyright (C) 2005-2006 Linus Torvalds <torvalds@osdl.org>
17 * Copyright (C) 2006 Ingo Molnar <mingo@elte.hu>
Jens Axboe5274f052006-03-30 15:15:30 +020018 *
19 */
20#include <linux/fs.h>
21#include <linux/file.h>
22#include <linux/pagemap.h>
Jens Axboed6b29d72007-06-04 09:59:47 +020023#include <linux/splice.h>
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -080024#include <linux/memcontrol.h>
Jens Axboe5274f052006-03-30 15:15:30 +020025#include <linux/mm_inline.h>
Jens Axboe5abc97a2006-03-30 15:16:46 +020026#include <linux/swap.h>
Jens Axboe4f6f0bd2006-04-02 23:04:46 +020027#include <linux/writeback.h>
Paul Gortmaker630d9c42011-11-16 23:57:37 -050028#include <linux/export.h>
Jens Axboe4f6f0bd2006-04-02 23:04:46 +020029#include <linux/syscalls.h>
Jens Axboe912d35f2006-04-26 10:59:21 +020030#include <linux/uio.h>
James Morris29ce2052007-07-13 11:44:32 +020031#include <linux/security.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090032#include <linux/gfp.h>
Eric Dumazet35f9c092012-04-05 03:05:35 +000033#include <linux/socket.h>
Al Viro76b021d2013-03-02 10:19:56 -050034#include <linux/compat.h>
Al Viro06ae43f2013-03-20 13:19:30 -040035#include "internal.h"
Jens Axboe5274f052006-03-30 15:15:30 +020036
Jens Axboe83f91352006-04-02 23:05:09 +020037/*
38 * Attempt to steal a page from a pipe buffer. This should perhaps go into
39 * a vm helper function, it's already simplified quite a bit by the
40 * addition of remove_mapping(). If success is returned, the caller may
41 * attempt to reuse this page for another destination.
42 */
Jens Axboe76ad4d12006-05-03 10:41:33 +020043static int page_cache_pipe_buf_steal(struct pipe_inode_info *pipe,
Jens Axboe5abc97a2006-03-30 15:16:46 +020044 struct pipe_buffer *buf)
45{
46 struct page *page = buf->page;
Jens Axboe9e94cd42006-06-20 15:01:12 +020047 struct address_space *mapping;
Jens Axboe5abc97a2006-03-30 15:16:46 +020048
Jens Axboe9e0267c2006-04-19 15:57:31 +020049 lock_page(page);
50
Jens Axboe9e94cd42006-06-20 15:01:12 +020051 mapping = page_mapping(page);
52 if (mapping) {
53 WARN_ON(!PageUptodate(page));
Jens Axboe5abc97a2006-03-30 15:16:46 +020054
Jens Axboe9e94cd42006-06-20 15:01:12 +020055 /*
56 * At least for ext2 with nobh option, we need to wait on
57 * writeback completing on this page, since we'll remove it
58 * from the pagecache. Otherwise truncate wont wait on the
59 * page, allowing the disk blocks to be reused by someone else
60 * before we actually wrote our data to them. fs corruption
61 * ensues.
62 */
63 wait_on_page_writeback(page);
Jens Axboead8d6f02006-04-02 23:10:32 +020064
David Howells266cf652009-04-03 16:42:36 +010065 if (page_has_private(page) &&
66 !try_to_release_page(page, GFP_KERNEL))
Jens Axboeca39d652008-05-20 21:27:41 +020067 goto out_unlock;
Jens Axboe4f6f0bd2006-04-02 23:04:46 +020068
Jens Axboe9e94cd42006-06-20 15:01:12 +020069 /*
70 * If we succeeded in removing the mapping, set LRU flag
71 * and return good.
72 */
73 if (remove_mapping(mapping, page)) {
74 buf->flags |= PIPE_BUF_FLAG_LRU;
75 return 0;
76 }
Jens Axboe9e0267c2006-04-19 15:57:31 +020077 }
Jens Axboe5abc97a2006-03-30 15:16:46 +020078
Jens Axboe9e94cd42006-06-20 15:01:12 +020079 /*
80 * Raced with truncate or failed to remove page from current
81 * address space, unlock and return failure.
82 */
Jens Axboeca39d652008-05-20 21:27:41 +020083out_unlock:
Jens Axboe9e94cd42006-06-20 15:01:12 +020084 unlock_page(page);
85 return 1;
Jens Axboe5abc97a2006-03-30 15:16:46 +020086}
87
Jens Axboe76ad4d12006-05-03 10:41:33 +020088static void page_cache_pipe_buf_release(struct pipe_inode_info *pipe,
Jens Axboe5274f052006-03-30 15:15:30 +020089 struct pipe_buffer *buf)
90{
91 page_cache_release(buf->page);
Jens Axboe14328732006-05-03 10:35:26 +020092 buf->flags &= ~PIPE_BUF_FLAG_LRU;
Jens Axboe5274f052006-03-30 15:15:30 +020093}
94
Jens Axboe08457182007-06-12 20:51:32 +020095/*
96 * Check whether the contents of buf is OK to access. Since the content
97 * is a page cache page, IO may be in flight.
98 */
Jens Axboecac36bb02007-06-14 13:10:48 +020099static int page_cache_pipe_buf_confirm(struct pipe_inode_info *pipe,
100 struct pipe_buffer *buf)
Jens Axboe5274f052006-03-30 15:15:30 +0200101{
102 struct page *page = buf->page;
Jens Axboe49d0b212006-04-10 09:04:41 +0200103 int err;
Jens Axboe5274f052006-03-30 15:15:30 +0200104
105 if (!PageUptodate(page)) {
Jens Axboe49d0b212006-04-10 09:04:41 +0200106 lock_page(page);
107
108 /*
109 * Page got truncated/unhashed. This will cause a 0-byte
Ingo Molnar73d62d82006-04-11 13:57:21 +0200110 * splice, if this is the first page.
Jens Axboe49d0b212006-04-10 09:04:41 +0200111 */
112 if (!page->mapping) {
113 err = -ENODATA;
114 goto error;
115 }
116
117 /*
Ingo Molnar73d62d82006-04-11 13:57:21 +0200118 * Uh oh, read-error from disk.
Jens Axboe49d0b212006-04-10 09:04:41 +0200119 */
120 if (!PageUptodate(page)) {
121 err = -EIO;
122 goto error;
123 }
124
125 /*
Jens Axboef84d7512006-05-01 19:59:03 +0200126 * Page is ok afterall, we are done.
Jens Axboe49d0b212006-04-10 09:04:41 +0200127 */
Jens Axboe5274f052006-03-30 15:15:30 +0200128 unlock_page(page);
Jens Axboe5274f052006-03-30 15:15:30 +0200129 }
130
Jens Axboef84d7512006-05-01 19:59:03 +0200131 return 0;
Jens Axboe49d0b212006-04-10 09:04:41 +0200132error:
133 unlock_page(page);
Jens Axboef84d7512006-05-01 19:59:03 +0200134 return err;
Jens Axboe70524492006-04-11 15:51:17 +0200135}
136
Hugh Dickins708e3502011-07-25 17:12:32 -0700137const struct pipe_buf_operations page_cache_pipe_buf_ops = {
Jens Axboe5274f052006-03-30 15:15:30 +0200138 .can_merge = 0,
Jens Axboecac36bb02007-06-14 13:10:48 +0200139 .confirm = page_cache_pipe_buf_confirm,
Jens Axboe5274f052006-03-30 15:15:30 +0200140 .release = page_cache_pipe_buf_release,
Jens Axboe5abc97a2006-03-30 15:16:46 +0200141 .steal = page_cache_pipe_buf_steal,
Jens Axboef84d7512006-05-01 19:59:03 +0200142 .get = generic_pipe_buf_get,
Jens Axboe5274f052006-03-30 15:15:30 +0200143};
144
Jens Axboe912d35f2006-04-26 10:59:21 +0200145static int user_page_pipe_buf_steal(struct pipe_inode_info *pipe,
146 struct pipe_buffer *buf)
147{
Jens Axboe7afa6fd2006-05-01 20:02:33 +0200148 if (!(buf->flags & PIPE_BUF_FLAG_GIFT))
149 return 1;
150
Jens Axboe14328732006-05-03 10:35:26 +0200151 buf->flags |= PIPE_BUF_FLAG_LRU;
Jens Axboe330ab712006-05-02 15:29:57 +0200152 return generic_pipe_buf_steal(pipe, buf);
Jens Axboe912d35f2006-04-26 10:59:21 +0200153}
154
Eric Dumazetd4c3cca2006-12-13 00:34:04 -0800155static const struct pipe_buf_operations user_page_pipe_buf_ops = {
Jens Axboe912d35f2006-04-26 10:59:21 +0200156 .can_merge = 0,
Jens Axboecac36bb02007-06-14 13:10:48 +0200157 .confirm = generic_pipe_buf_confirm,
Jens Axboe912d35f2006-04-26 10:59:21 +0200158 .release = page_cache_pipe_buf_release,
159 .steal = user_page_pipe_buf_steal,
Jens Axboef84d7512006-05-01 19:59:03 +0200160 .get = generic_pipe_buf_get,
Jens Axboe912d35f2006-04-26 10:59:21 +0200161};
162
Namhyung Kim825cdcb2011-05-23 19:58:53 +0200163static void wakeup_pipe_readers(struct pipe_inode_info *pipe)
164{
165 smp_mb();
166 if (waitqueue_active(&pipe->wait))
167 wake_up_interruptible(&pipe->wait);
168 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
169}
170
Jens Axboe932cc6d2007-06-21 13:10:21 +0200171/**
172 * splice_to_pipe - fill passed data into a pipe
173 * @pipe: pipe to fill
174 * @spd: data to fill
175 *
176 * Description:
Randy Dunlap79685b82007-07-27 08:08:51 +0200177 * @spd contains a map of pages and len/offset tuples, along with
Jens Axboe932cc6d2007-06-21 13:10:21 +0200178 * the struct pipe_buf_operations associated with these pages. This
179 * function will link that data to the pipe.
180 *
Jens Axboe83f91352006-04-02 23:05:09 +0200181 */
Jens Axboed6b29d72007-06-04 09:59:47 +0200182ssize_t splice_to_pipe(struct pipe_inode_info *pipe,
183 struct splice_pipe_desc *spd)
Jens Axboe5274f052006-03-30 15:15:30 +0200184{
Jens Axboe00de00b2007-06-15 13:14:22 +0200185 unsigned int spd_pages = spd->nr_pages;
Jens Axboe912d35f2006-04-26 10:59:21 +0200186 int ret, do_wakeup, page_nr;
Jens Axboe5274f052006-03-30 15:15:30 +0200187
Rabin Vincentff6c4182016-03-10 21:19:06 +0100188 if (!spd_pages)
189 return 0;
190
Jens Axboe5274f052006-03-30 15:15:30 +0200191 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;
Miklos Szeredi4cc0d2f2017-02-16 17:49:02 +0100214 buf->flags = 0;
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 Wu431a4822007-07-19 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 Wu431a4822007-07-19 01:48:05 -0700344 * readahead/allocate the rest and fill in the holes.
Jens Axboeeb207962006-04-27 11:05:22 +0200345 */
Fengguang Wu431a4822007-07-19 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 Wu431a4822007-07-19 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
Boaz Harroshbe64f882015-04-15 16:15:17 -0700530 if (IS_DAX(in->f_mapping->host))
531 return default_file_splice_read(in, ppos, pipe, len, flags);
532
Jens Axboed366d3982007-06-01 14:54:11 +0200533 isize = i_size_read(in->f_mapping->host);
534 if (unlikely(*ppos >= isize))
535 return 0;
536
537 left = isize - *ppos;
538 if (unlikely(left < len))
539 len = left;
Jens Axboe5274f052006-03-30 15:15:30 +0200540
Jens Axboe8191ecd2008-04-10 08:24:25 +0200541 ret = __generic_file_splice_read(in, ppos, pipe, len, flags);
Miklos Szeredi723590e2009-08-15 08:43:22 +0200542 if (ret > 0) {
Jens Axboecbb7e572006-04-11 14:57:50 +0200543 *ppos += ret;
Miklos Szeredi723590e2009-08-15 08:43:22 +0200544 file_accessed(in);
545 }
Jens Axboe5274f052006-03-30 15:15:30 +0200546
547 return ret;
548}
Jens Axboe059a8f32006-04-02 23:06:05 +0200549EXPORT_SYMBOL(generic_file_splice_read);
550
Miklos Szeredi68181732009-05-07 15:37:36 +0200551static const struct pipe_buf_operations default_pipe_buf_ops = {
552 .can_merge = 0,
Miklos Szeredi68181732009-05-07 15:37:36 +0200553 .confirm = generic_pipe_buf_confirm,
554 .release = generic_pipe_buf_release,
555 .steal = generic_pipe_buf_steal,
556 .get = generic_pipe_buf_get,
557};
558
Miklos Szeredi28a625c2014-01-22 19:36:57 +0100559static int generic_pipe_buf_nosteal(struct pipe_inode_info *pipe,
560 struct pipe_buffer *buf)
561{
562 return 1;
563}
564
565/* Pipe buffer operations for a socket and similar. */
566const struct pipe_buf_operations nosteal_pipe_buf_ops = {
567 .can_merge = 0,
Miklos Szeredi28a625c2014-01-22 19:36:57 +0100568 .confirm = generic_pipe_buf_confirm,
569 .release = generic_pipe_buf_release,
570 .steal = generic_pipe_buf_nosteal,
571 .get = generic_pipe_buf_get,
572};
573EXPORT_SYMBOL(nosteal_pipe_buf_ops);
574
Miklos Szeredi68181732009-05-07 15:37:36 +0200575static ssize_t kernel_readv(struct file *file, const struct iovec *vec,
576 unsigned long vlen, loff_t offset)
577{
578 mm_segment_t old_fs;
579 loff_t pos = offset;
580 ssize_t res;
581
582 old_fs = get_fs();
583 set_fs(get_ds());
584 /* The cast to a user pointer is valid due to the set_fs() */
585 res = vfs_readv(file, (const struct iovec __user *)vec, vlen, &pos);
586 set_fs(old_fs);
587
588 return res;
589}
590
Al Viro7bb307e2013-02-23 14:51:48 -0500591ssize_t kernel_write(struct file *file, const char *buf, size_t count,
Miklos Szeredib2858d72009-05-19 11:37:46 +0200592 loff_t pos)
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +0200593{
594 mm_segment_t old_fs;
595 ssize_t res;
596
597 old_fs = get_fs();
598 set_fs(get_ds());
599 /* The cast to a user pointer is valid due to the set_fs() */
Al Viro7bb307e2013-02-23 14:51:48 -0500600 res = vfs_write(file, (__force const char __user *)buf, count, &pos);
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +0200601 set_fs(old_fs);
602
603 return res;
604}
Al Viro7bb307e2013-02-23 14:51:48 -0500605EXPORT_SYMBOL(kernel_write);
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +0200606
Miklos Szeredi68181732009-05-07 15:37:36 +0200607ssize_t default_file_splice_read(struct file *in, loff_t *ppos,
608 struct pipe_inode_info *pipe, size_t len,
609 unsigned int flags)
610{
611 unsigned int nr_pages;
612 unsigned int nr_freed;
613 size_t offset;
Jens Axboe35f3d142010-05-20 10:43:18 +0200614 struct page *pages[PIPE_DEF_BUFFERS];
615 struct partial_page partial[PIPE_DEF_BUFFERS];
616 struct iovec *vec, __vec[PIPE_DEF_BUFFERS];
Miklos Szeredi68181732009-05-07 15:37:36 +0200617 ssize_t res;
618 size_t this_len;
619 int error;
620 int i;
621 struct splice_pipe_desc spd = {
622 .pages = pages,
623 .partial = partial,
Eric Dumazet047fe362012-06-12 15:24:40 +0200624 .nr_pages_max = PIPE_DEF_BUFFERS,
Miklos Szeredi68181732009-05-07 15:37:36 +0200625 .flags = flags,
626 .ops = &default_pipe_buf_ops,
627 .spd_release = spd_release_page,
628 };
629
Jens Axboe35f3d142010-05-20 10:43:18 +0200630 if (splice_grow_spd(pipe, &spd))
631 return -ENOMEM;
632
633 res = -ENOMEM;
634 vec = __vec;
Eric Dumazet047fe362012-06-12 15:24:40 +0200635 if (spd.nr_pages_max > PIPE_DEF_BUFFERS) {
636 vec = kmalloc(spd.nr_pages_max * sizeof(struct iovec), GFP_KERNEL);
Jens Axboe35f3d142010-05-20 10:43:18 +0200637 if (!vec)
638 goto shrink_ret;
639 }
640
Miklos Szeredi68181732009-05-07 15:37:36 +0200641 offset = *ppos & ~PAGE_CACHE_MASK;
642 nr_pages = (len + offset + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
643
Eric Dumazet047fe362012-06-12 15:24:40 +0200644 for (i = 0; i < nr_pages && i < spd.nr_pages_max && len; i++) {
Miklos Szeredi68181732009-05-07 15:37:36 +0200645 struct page *page;
646
Jens Axboe4f231222009-05-13 08:35:35 +0200647 page = alloc_page(GFP_USER);
Miklos Szeredi68181732009-05-07 15:37:36 +0200648 error = -ENOMEM;
649 if (!page)
650 goto err;
651
652 this_len = min_t(size_t, len, PAGE_CACHE_SIZE - offset);
Jens Axboe4f231222009-05-13 08:35:35 +0200653 vec[i].iov_base = (void __user *) page_address(page);
Miklos Szeredi68181732009-05-07 15:37:36 +0200654 vec[i].iov_len = this_len;
Jens Axboe35f3d142010-05-20 10:43:18 +0200655 spd.pages[i] = page;
Miklos Szeredi68181732009-05-07 15:37:36 +0200656 spd.nr_pages++;
657 len -= this_len;
658 offset = 0;
659 }
660
661 res = kernel_readv(in, vec, spd.nr_pages, *ppos);
Andrew Morton77f6bf52009-05-14 09:49:44 +0200662 if (res < 0) {
663 error = res;
Miklos Szeredi68181732009-05-07 15:37:36 +0200664 goto err;
Andrew Morton77f6bf52009-05-14 09:49:44 +0200665 }
Miklos Szeredi68181732009-05-07 15:37:36 +0200666
667 error = 0;
668 if (!res)
669 goto err;
670
671 nr_freed = 0;
672 for (i = 0; i < spd.nr_pages; i++) {
Miklos Szeredi68181732009-05-07 15:37:36 +0200673 this_len = min_t(size_t, vec[i].iov_len, res);
Jens Axboe35f3d142010-05-20 10:43:18 +0200674 spd.partial[i].offset = 0;
675 spd.partial[i].len = this_len;
Miklos Szeredi68181732009-05-07 15:37:36 +0200676 if (!this_len) {
Jens Axboe35f3d142010-05-20 10:43:18 +0200677 __free_page(spd.pages[i]);
678 spd.pages[i] = NULL;
Miklos Szeredi68181732009-05-07 15:37:36 +0200679 nr_freed++;
680 }
681 res -= this_len;
682 }
683 spd.nr_pages -= nr_freed;
684
685 res = splice_to_pipe(pipe, &spd);
686 if (res > 0)
687 *ppos += res;
688
Jens Axboe35f3d142010-05-20 10:43:18 +0200689shrink_ret:
690 if (vec != __vec)
691 kfree(vec);
Eric Dumazet047fe362012-06-12 15:24:40 +0200692 splice_shrink_spd(&spd);
Miklos Szeredi68181732009-05-07 15:37:36 +0200693 return res;
694
695err:
Jens Axboe4f231222009-05-13 08:35:35 +0200696 for (i = 0; i < spd.nr_pages; i++)
Jens Axboe35f3d142010-05-20 10:43:18 +0200697 __free_page(spd.pages[i]);
Jens Axboe4f231222009-05-13 08:35:35 +0200698
Jens Axboe35f3d142010-05-20 10:43:18 +0200699 res = error;
700 goto shrink_ret;
Miklos Szeredi68181732009-05-07 15:37:36 +0200701}
702EXPORT_SYMBOL(default_file_splice_read);
703
Jens Axboe5274f052006-03-30 15:15:30 +0200704/*
Jens Axboe4f6f0bd2006-04-02 23:04:46 +0200705 * Send 'sd->len' bytes to socket from 'sd->file' at position 'sd->pos'
Jens Axboe016b6612006-04-25 15:42:00 +0200706 * using sendpage(). Return the number of bytes sent.
Jens Axboe5274f052006-03-30 15:15:30 +0200707 */
Jens Axboe76ad4d12006-05-03 10:41:33 +0200708static int pipe_to_sendpage(struct pipe_inode_info *pipe,
Jens Axboe5274f052006-03-30 15:15:30 +0200709 struct pipe_buffer *buf, struct splice_desc *sd)
710{
Jens Axboe6a14b902007-06-14 13:08:55 +0200711 struct file *file = sd->u.file;
Jens Axboe5274f052006-03-30 15:15:30 +0200712 loff_t pos = sd->pos;
Michał Mirosława8adbe32010-12-17 08:56:44 +0100713 int more;
Jens Axboe5274f052006-03-30 15:15:30 +0200714
Al Viro72c2d532013-09-22 16:27:52 -0400715 if (!likely(file->f_op->sendpage))
Michał Mirosława8adbe32010-12-17 08:56:44 +0100716 return -EINVAL;
Jens Axboe5274f052006-03-30 15:15:30 +0200717
Eric Dumazet35f9c092012-04-05 03:05:35 +0000718 more = (sd->flags & SPLICE_F_MORE) ? MSG_MORE : 0;
Eric Dumazetae62ca72013-01-06 18:21:49 +0000719
720 if (sd->len < sd->total_len && pipe->nrbufs > 1)
Eric Dumazet35f9c092012-04-05 03:05:35 +0000721 more |= MSG_SENDPAGE_NOTLAST;
Eric Dumazetae62ca72013-01-06 18:21:49 +0000722
Michał Mirosława8adbe32010-12-17 08:56:44 +0100723 return file->f_op->sendpage(file, buf->page, buf->offset,
724 sd->len, &pos, more);
Jens Axboe5274f052006-03-30 15:15:30 +0200725}
726
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200727static void wakeup_pipe_writers(struct pipe_inode_info *pipe)
728{
729 smp_mb();
730 if (waitqueue_active(&pipe->wait))
731 wake_up_interruptible(&pipe->wait);
732 kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
733}
734
735/**
736 * splice_from_pipe_feed - feed available data from a pipe to a file
737 * @pipe: pipe to splice from
738 * @sd: information to @actor
739 * @actor: handler that splices the data
740 *
741 * Description:
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200742 * This function loops over the pipe and calls @actor to do the
743 * actual moving of a single struct pipe_buffer to the desired
744 * destination. It returns when there's no more buffers left in
745 * the pipe or if the requested number of bytes (@sd->total_len)
746 * have been copied. It returns a positive number (one) if the
747 * pipe needs to be filled with more data, zero if the required
748 * number of bytes have been copied and -errno on error.
749 *
750 * This, together with splice_from_pipe_{begin,end,next}, may be
751 * used to implement the functionality of __splice_from_pipe() when
752 * locking is required around copying the pipe buffers to the
753 * destination.
754 */
Al Viro96f9bc82014-04-05 04:35:49 -0400755static int splice_from_pipe_feed(struct pipe_inode_info *pipe, struct splice_desc *sd,
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200756 splice_actor *actor)
757{
758 int ret;
759
760 while (pipe->nrbufs) {
761 struct pipe_buffer *buf = pipe->bufs + pipe->curbuf;
762 const struct pipe_buf_operations *ops = buf->ops;
763
764 sd->len = buf->len;
765 if (sd->len > sd->total_len)
766 sd->len = sd->total_len;
767
Michał Mirosława8adbe32010-12-17 08:56:44 +0100768 ret = buf->ops->confirm(pipe, buf);
769 if (unlikely(ret)) {
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200770 if (ret == -ENODATA)
771 ret = 0;
772 return ret;
773 }
Michał Mirosława8adbe32010-12-17 08:56:44 +0100774
775 ret = actor(pipe, buf, sd);
776 if (ret <= 0)
777 return ret;
778
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200779 buf->offset += ret;
780 buf->len -= ret;
781
782 sd->num_spliced += ret;
783 sd->len -= ret;
784 sd->pos += ret;
785 sd->total_len -= ret;
786
787 if (!buf->len) {
788 buf->ops = NULL;
789 ops->release(pipe, buf);
Jens Axboe35f3d142010-05-20 10:43:18 +0200790 pipe->curbuf = (pipe->curbuf + 1) & (pipe->buffers - 1);
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200791 pipe->nrbufs--;
Al Viro6447a3c2013-03-21 11:01:38 -0400792 if (pipe->files)
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200793 sd->need_wakeup = true;
794 }
795
796 if (!sd->total_len)
797 return 0;
798 }
799
800 return 1;
801}
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200802
803/**
804 * splice_from_pipe_next - wait for some data to splice from
805 * @pipe: pipe to splice from
806 * @sd: information about the splice operation
807 *
808 * Description:
809 * This function will wait for some data and return a positive
810 * value (one) if pipe buffers are available. It will return zero
811 * or -errno if no more data needs to be spliced.
812 */
Al Viro96f9bc82014-04-05 04:35:49 -0400813static int splice_from_pipe_next(struct pipe_inode_info *pipe, struct splice_desc *sd)
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200814{
815 while (!pipe->nrbufs) {
816 if (!pipe->writers)
817 return 0;
818
819 if (!pipe->waiting_writers && sd->num_spliced)
820 return 0;
821
822 if (sd->flags & SPLICE_F_NONBLOCK)
823 return -EAGAIN;
824
825 if (signal_pending(current))
826 return -ERESTARTSYS;
827
828 if (sd->need_wakeup) {
829 wakeup_pipe_writers(pipe);
830 sd->need_wakeup = false;
831 }
832
833 pipe_wait(pipe);
834 }
835
836 return 1;
837}
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200838
839/**
840 * splice_from_pipe_begin - start splicing from pipe
Randy Dunlapb80901b2009-04-16 19:09:55 -0700841 * @sd: information about the splice operation
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200842 *
843 * Description:
844 * This function should be called before a loop containing
845 * splice_from_pipe_next() and splice_from_pipe_feed() to
846 * initialize the necessary fields of @sd.
847 */
Al Viro96f9bc82014-04-05 04:35:49 -0400848static void splice_from_pipe_begin(struct splice_desc *sd)
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200849{
850 sd->num_spliced = 0;
851 sd->need_wakeup = false;
852}
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200853
854/**
855 * splice_from_pipe_end - finish splicing from pipe
856 * @pipe: pipe to splice from
857 * @sd: information about the splice operation
858 *
859 * Description:
860 * This function will wake up pipe writers if necessary. It should
861 * be called after a loop containing splice_from_pipe_next() and
862 * splice_from_pipe_feed().
863 */
Al Viro96f9bc82014-04-05 04:35:49 -0400864static void splice_from_pipe_end(struct pipe_inode_info *pipe, struct splice_desc *sd)
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200865{
866 if (sd->need_wakeup)
867 wakeup_pipe_writers(pipe);
868}
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200869
Jens Axboe932cc6d2007-06-21 13:10:21 +0200870/**
871 * __splice_from_pipe - splice data from a pipe to given actor
872 * @pipe: pipe to splice from
873 * @sd: information to @actor
874 * @actor: handler that splices the data
875 *
876 * Description:
877 * This function does little more than loop over the pipe and call
878 * @actor to do the actual moving of a single struct pipe_buffer to
879 * the desired destination. See pipe_to_file, pipe_to_sendpage, or
880 * pipe_to_user.
881 *
Jens Axboe83f91352006-04-02 23:05:09 +0200882 */
Jens Axboec66ab6f2007-06-12 21:17:17 +0200883ssize_t __splice_from_pipe(struct pipe_inode_info *pipe, struct splice_desc *sd,
884 splice_actor *actor)
Jens Axboe5274f052006-03-30 15:15:30 +0200885{
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200886 int ret;
Jens Axboe5274f052006-03-30 15:15:30 +0200887
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200888 splice_from_pipe_begin(sd);
889 do {
890 ret = splice_from_pipe_next(pipe, sd);
891 if (ret > 0)
892 ret = splice_from_pipe_feed(pipe, sd, actor);
893 } while (ret > 0);
894 splice_from_pipe_end(pipe, sd);
Jens Axboe5274f052006-03-30 15:15:30 +0200895
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200896 return sd->num_spliced ? sd->num_spliced : ret;
Jens Axboe5274f052006-03-30 15:15:30 +0200897}
Mark Fasheh40bee44e2007-03-21 13:11:02 +0100898EXPORT_SYMBOL(__splice_from_pipe);
Jens Axboe5274f052006-03-30 15:15:30 +0200899
Jens Axboe932cc6d2007-06-21 13:10:21 +0200900/**
901 * splice_from_pipe - splice data from a pipe to a file
902 * @pipe: pipe to splice from
903 * @out: file to splice to
904 * @ppos: position in @out
905 * @len: how many bytes to splice
906 * @flags: splice modifier flags
907 * @actor: handler that splices the data
908 *
909 * Description:
Miklos Szeredi29339702009-04-14 19:48:37 +0200910 * See __splice_from_pipe. This function locks the pipe inode,
Jens Axboe932cc6d2007-06-21 13:10:21 +0200911 * otherwise it's identical to __splice_from_pipe().
912 *
913 */
Mark Fasheh6da61802006-10-17 18:43:07 +0200914ssize_t splice_from_pipe(struct pipe_inode_info *pipe, struct file *out,
915 loff_t *ppos, size_t len, unsigned int flags,
916 splice_actor *actor)
917{
918 ssize_t ret;
Jens Axboec66ab6f2007-06-12 21:17:17 +0200919 struct splice_desc sd = {
920 .total_len = len,
921 .flags = flags,
922 .pos = *ppos,
Jens Axboe6a14b902007-06-14 13:08:55 +0200923 .u.file = out,
Jens Axboec66ab6f2007-06-12 21:17:17 +0200924 };
Mark Fasheh6da61802006-10-17 18:43:07 +0200925
Miklos Szeredi61e0d472009-04-14 19:48:41 +0200926 pipe_lock(pipe);
Jens Axboec66ab6f2007-06-12 21:17:17 +0200927 ret = __splice_from_pipe(pipe, &sd, actor);
Miklos Szeredi61e0d472009-04-14 19:48:41 +0200928 pipe_unlock(pipe);
Mark Fasheh6da61802006-10-17 18:43:07 +0200929
930 return ret;
931}
932
933/**
Al Viro8d020762014-04-05 04:27:08 -0400934 * iter_file_splice_write - splice data from a pipe to a file
935 * @pipe: pipe info
936 * @out: file to write to
937 * @ppos: position in @out
938 * @len: number of bytes to splice
939 * @flags: splice modifier flags
940 *
941 * Description:
942 * Will either move or copy pages (determined by @flags options) from
943 * the given pipe inode to the given file.
944 * This one is ->write_iter-based.
945 *
946 */
947ssize_t
948iter_file_splice_write(struct pipe_inode_info *pipe, struct file *out,
949 loff_t *ppos, size_t len, unsigned int flags)
950{
951 struct splice_desc sd = {
952 .total_len = len,
953 .flags = flags,
954 .pos = *ppos,
955 .u.file = out,
956 };
957 int nbufs = pipe->buffers;
958 struct bio_vec *array = kcalloc(nbufs, sizeof(struct bio_vec),
959 GFP_KERNEL);
960 ssize_t ret;
961
962 if (unlikely(!array))
963 return -ENOMEM;
964
965 pipe_lock(pipe);
966
967 splice_from_pipe_begin(&sd);
968 while (sd.total_len) {
969 struct iov_iter from;
Al Viro8d020762014-04-05 04:27:08 -0400970 size_t left;
971 int n, idx;
972
973 ret = splice_from_pipe_next(pipe, &sd);
974 if (ret <= 0)
975 break;
976
977 if (unlikely(nbufs < pipe->buffers)) {
978 kfree(array);
979 nbufs = pipe->buffers;
980 array = kcalloc(nbufs, sizeof(struct bio_vec),
981 GFP_KERNEL);
982 if (!array) {
983 ret = -ENOMEM;
984 break;
985 }
986 }
987
988 /* build the vector */
989 left = sd.total_len;
990 for (n = 0, idx = pipe->curbuf; left && n < pipe->nrbufs; n++, idx++) {
991 struct pipe_buffer *buf = pipe->bufs + idx;
992 size_t this_len = buf->len;
993
994 if (this_len > left)
995 this_len = left;
996
997 if (idx == pipe->buffers - 1)
998 idx = -1;
999
1000 ret = buf->ops->confirm(pipe, buf);
1001 if (unlikely(ret)) {
1002 if (ret == -ENODATA)
1003 ret = 0;
1004 goto done;
1005 }
1006
1007 array[n].bv_page = buf->page;
1008 array[n].bv_len = this_len;
1009 array[n].bv_offset = buf->offset;
1010 left -= this_len;
1011 }
1012
Al Viro05afcb72015-01-23 01:08:07 -05001013 iov_iter_bvec(&from, ITER_BVEC | WRITE, array, n,
1014 sd.total_len - left);
Christoph Hellwigdbe4e192015-01-25 21:11:59 +01001015 ret = vfs_iter_write(out, &from, &sd.pos);
Al Viro8d020762014-04-05 04:27:08 -04001016 if (ret <= 0)
1017 break;
1018
1019 sd.num_spliced += ret;
1020 sd.total_len -= ret;
Christoph Hellwigdbe4e192015-01-25 21:11:59 +01001021 *ppos = sd.pos;
Al Viro8d020762014-04-05 04:27:08 -04001022
1023 /* dismiss the fully eaten buffers, adjust the partial one */
1024 while (ret) {
1025 struct pipe_buffer *buf = pipe->bufs + pipe->curbuf;
1026 if (ret >= buf->len) {
1027 const struct pipe_buf_operations *ops = buf->ops;
1028 ret -= buf->len;
1029 buf->len = 0;
1030 buf->ops = NULL;
1031 ops->release(pipe, buf);
1032 pipe->curbuf = (pipe->curbuf + 1) & (pipe->buffers - 1);
1033 pipe->nrbufs--;
1034 if (pipe->files)
1035 sd.need_wakeup = true;
1036 } else {
1037 buf->offset += ret;
1038 buf->len -= ret;
1039 ret = 0;
1040 }
1041 }
1042 }
1043done:
1044 kfree(array);
1045 splice_from_pipe_end(pipe, &sd);
1046
1047 pipe_unlock(pipe);
1048
1049 if (sd.num_spliced)
1050 ret = sd.num_spliced;
1051
1052 return ret;
1053}
1054
1055EXPORT_SYMBOL(iter_file_splice_write);
1056
Miklos Szeredib2858d72009-05-19 11:37:46 +02001057static int write_pipe_buf(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
1058 struct splice_desc *sd)
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +02001059{
Miklos Szeredib2858d72009-05-19 11:37:46 +02001060 int ret;
1061 void *data;
Al Viro06ae43f2013-03-20 13:19:30 -04001062 loff_t tmp = sd->pos;
Miklos Szeredib2858d72009-05-19 11:37:46 +02001063
Al Virofbb32752014-02-02 21:09:54 -05001064 data = kmap(buf->page);
Al Viro06ae43f2013-03-20 13:19:30 -04001065 ret = __kernel_write(sd->u.file, data + buf->offset, sd->len, &tmp);
Al Virofbb32752014-02-02 21:09:54 -05001066 kunmap(buf->page);
Miklos Szeredib2858d72009-05-19 11:37:46 +02001067
1068 return ret;
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +02001069}
1070
1071static ssize_t default_file_splice_write(struct pipe_inode_info *pipe,
1072 struct file *out, loff_t *ppos,
1073 size_t len, unsigned int flags)
1074{
Miklos Szeredib2858d72009-05-19 11:37:46 +02001075 ssize_t ret;
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +02001076
Miklos Szeredib2858d72009-05-19 11:37:46 +02001077 ret = splice_from_pipe(pipe, out, ppos, len, flags, write_pipe_buf);
1078 if (ret > 0)
1079 *ppos += ret;
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +02001080
Miklos Szeredib2858d72009-05-19 11:37:46 +02001081 return ret;
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +02001082}
1083
Jens Axboe83f91352006-04-02 23:05:09 +02001084/**
1085 * generic_splice_sendpage - splice data from a pipe to a socket
Jens Axboe932cc6d2007-06-21 13:10:21 +02001086 * @pipe: pipe to splice from
Jens Axboe83f91352006-04-02 23:05:09 +02001087 * @out: socket to write to
Jens Axboe932cc6d2007-06-21 13:10:21 +02001088 * @ppos: position in @out
Jens Axboe83f91352006-04-02 23:05:09 +02001089 * @len: number of bytes to splice
1090 * @flags: splice modifier flags
1091 *
Jens Axboe932cc6d2007-06-21 13:10:21 +02001092 * Description:
1093 * Will send @len bytes from the pipe to a network socket. No data copying
1094 * is involved.
Jens Axboe83f91352006-04-02 23:05:09 +02001095 *
1096 */
Ingo Molnar3a326a22006-04-10 15:18:35 +02001097ssize_t generic_splice_sendpage(struct pipe_inode_info *pipe, struct file *out,
Jens Axboecbb7e572006-04-11 14:57:50 +02001098 loff_t *ppos, size_t len, unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +02001099{
Jens Axboe00522fb2006-04-26 14:39:29 +02001100 return splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_sendpage);
Jens Axboe5274f052006-03-30 15:15:30 +02001101}
1102
Jens Axboe059a8f32006-04-02 23:06:05 +02001103EXPORT_SYMBOL(generic_splice_sendpage);
Jeff Garzika0f06782006-03-30 23:06:13 -05001104
Jens Axboe83f91352006-04-02 23:05:09 +02001105/*
1106 * Attempt to initiate a splice from pipe to file.
1107 */
Ingo Molnar3a326a22006-04-10 15:18:35 +02001108static long do_splice_from(struct pipe_inode_info *pipe, struct file *out,
Jens Axboecbb7e572006-04-11 14:57:50 +02001109 loff_t *ppos, size_t len, unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +02001110{
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +02001111 ssize_t (*splice_write)(struct pipe_inode_info *, struct file *,
1112 loff_t *, size_t, unsigned int);
Jens Axboe5274f052006-03-30 15:15:30 +02001113
Al Viro72c2d532013-09-22 16:27:52 -04001114 if (out->f_op->splice_write)
Changli Gaocc56f7d2009-11-04 09:09:52 +01001115 splice_write = out->f_op->splice_write;
1116 else
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +02001117 splice_write = default_file_splice_write;
1118
Al Viro500368f2013-05-23 20:07:11 -04001119 return splice_write(pipe, out, ppos, len, flags);
Jens Axboe5274f052006-03-30 15:15:30 +02001120}
1121
Jens Axboe83f91352006-04-02 23:05:09 +02001122/*
1123 * Attempt to initiate a splice from a file to a pipe.
1124 */
Jens Axboecbb7e572006-04-11 14:57:50 +02001125static long do_splice_to(struct file *in, loff_t *ppos,
1126 struct pipe_inode_info *pipe, size_t len,
1127 unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +02001128{
Miklos Szeredi68181732009-05-07 15:37:36 +02001129 ssize_t (*splice_read)(struct file *, loff_t *,
1130 struct pipe_inode_info *, size_t, unsigned int);
Jens Axboe5274f052006-03-30 15:15:30 +02001131 int ret;
1132
Jens Axboe49570e92006-04-11 13:56:09 +02001133 if (unlikely(!(in->f_mode & FMODE_READ)))
Jens Axboe5274f052006-03-30 15:15:30 +02001134 return -EBADF;
1135
Jens Axboecbb7e572006-04-11 14:57:50 +02001136 ret = rw_verify_area(READ, in, ppos, len);
Jens Axboe5274f052006-03-30 15:15:30 +02001137 if (unlikely(ret < 0))
1138 return ret;
1139
Al Viro72c2d532013-09-22 16:27:52 -04001140 if (in->f_op->splice_read)
Changli Gaocc56f7d2009-11-04 09:09:52 +01001141 splice_read = in->f_op->splice_read;
1142 else
Miklos Szeredi68181732009-05-07 15:37:36 +02001143 splice_read = default_file_splice_read;
1144
1145 return splice_read(in, ppos, pipe, len, flags);
Jens Axboe5274f052006-03-30 15:15:30 +02001146}
1147
Jens Axboe932cc6d2007-06-21 13:10:21 +02001148/**
1149 * splice_direct_to_actor - splices data directly between two non-pipes
1150 * @in: file to splice from
1151 * @sd: actor information on where to splice to
1152 * @actor: handles the data splicing
1153 *
1154 * Description:
1155 * This is a special case helper to splice directly between two
1156 * points, without requiring an explicit pipe. Internally an allocated
Randy Dunlap79685b82007-07-27 08:08:51 +02001157 * pipe is cached in the process, and reused during the lifetime of
Jens Axboe932cc6d2007-06-21 13:10:21 +02001158 * that process.
1159 *
Jens Axboec66ab6f2007-06-12 21:17:17 +02001160 */
1161ssize_t splice_direct_to_actor(struct file *in, struct splice_desc *sd,
1162 splice_direct_actor *actor)
Jens Axboeb92ce552006-04-11 13:52:07 +02001163{
1164 struct pipe_inode_info *pipe;
1165 long ret, bytes;
1166 umode_t i_mode;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001167 size_t len;
Christophe Leroy0ff28d92015-05-06 17:26:47 +02001168 int i, flags, more;
Jens Axboeb92ce552006-04-11 13:52:07 +02001169
1170 /*
1171 * We require the input being a regular file, as we don't want to
1172 * randomly drop data for eg socket -> socket splicing. Use the
1173 * piped splicing for that!
1174 */
Al Viro496ad9a2013-01-23 17:07:38 -05001175 i_mode = file_inode(in)->i_mode;
Jens Axboeb92ce552006-04-11 13:52:07 +02001176 if (unlikely(!S_ISREG(i_mode) && !S_ISBLK(i_mode)))
1177 return -EINVAL;
1178
1179 /*
1180 * neither in nor out is a pipe, setup an internal pipe attached to
1181 * 'out' and transfer the wanted data from 'in' to 'out' through that
1182 */
1183 pipe = current->splice_pipe;
Jens Axboe49570e92006-04-11 13:56:09 +02001184 if (unlikely(!pipe)) {
Al Viro7bee1302013-03-21 11:04:15 -04001185 pipe = alloc_pipe_info();
Jens Axboeb92ce552006-04-11 13:52:07 +02001186 if (!pipe)
1187 return -ENOMEM;
1188
1189 /*
1190 * We don't have an immediate reader, but we'll read the stuff
Jens Axboe00522fb2006-04-26 14:39:29 +02001191 * out of the pipe right after the splice_to_pipe(). So set
Jens Axboeb92ce552006-04-11 13:52:07 +02001192 * PIPE_READERS appropriately.
1193 */
1194 pipe->readers = 1;
1195
1196 current->splice_pipe = pipe;
1197 }
1198
1199 /*
Ingo Molnar73d62d82006-04-11 13:57:21 +02001200 * Do the splice.
Jens Axboeb92ce552006-04-11 13:52:07 +02001201 */
1202 ret = 0;
1203 bytes = 0;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001204 len = sd->total_len;
1205 flags = sd->flags;
1206
1207 /*
1208 * Don't block on output, we have to drain the direct pipe.
1209 */
1210 sd->flags &= ~SPLICE_F_NONBLOCK;
Christophe Leroy0ff28d92015-05-06 17:26:47 +02001211 more = sd->flags & SPLICE_F_MORE;
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 /*
Christophe Leroy0ff28d92015-05-06 17:26:47 +02001225 * If more data is pending, set SPLICE_F_MORE
1226 * If this is the last data and SPLICE_F_MORE was not set
1227 * initially, clears it.
1228 */
1229 if (read_len < len)
1230 sd->flags |= SPLICE_F_MORE;
1231 else if (!more)
1232 sd->flags &= ~SPLICE_F_MORE;
1233 /*
Jens Axboeb92ce552006-04-11 13:52:07 +02001234 * NOTE: nonblocking mode only applies to the input. We
1235 * must not do the output in nonblocking mode as then we
1236 * could get stuck data in the internal pipe:
1237 */
Jens Axboec66ab6f2007-06-12 21:17:17 +02001238 ret = actor(pipe, sd);
Tom Zanussia82c53a2008-05-09 13:28:36 +02001239 if (unlikely(ret <= 0)) {
1240 sd->pos = prev_pos;
Jens Axboeb92ce552006-04-11 13:52:07 +02001241 goto out_release;
Tom Zanussia82c53a2008-05-09 13:28:36 +02001242 }
Jens Axboeb92ce552006-04-11 13:52:07 +02001243
1244 bytes += ret;
1245 len -= ret;
Jens Axboebcd4f3a2007-07-16 14:41:49 +02001246 sd->pos = pos;
Jens Axboeb92ce552006-04-11 13:52:07 +02001247
Tom Zanussia82c53a2008-05-09 13:28:36 +02001248 if (ret < read_len) {
1249 sd->pos = prev_pos + ret;
Jens Axboe51a92c02007-07-13 14:11:43 +02001250 goto out_release;
Tom Zanussia82c53a2008-05-09 13:28:36 +02001251 }
Jens Axboeb92ce552006-04-11 13:52:07 +02001252 }
1253
Jens Axboe9e971982008-01-29 21:05:57 +01001254done:
Jens Axboeb92ce552006-04-11 13:52:07 +02001255 pipe->nrbufs = pipe->curbuf = 0;
Jens Axboe80848702008-01-30 12:24:48 +01001256 file_accessed(in);
Jens Axboeb92ce552006-04-11 13:52:07 +02001257 return bytes;
1258
1259out_release:
1260 /*
1261 * If we did an incomplete transfer we must release
1262 * the pipe buffers in question:
1263 */
Jens Axboe35f3d142010-05-20 10:43:18 +02001264 for (i = 0; i < pipe->buffers; i++) {
Jens Axboeb92ce552006-04-11 13:52:07 +02001265 struct pipe_buffer *buf = pipe->bufs + i;
1266
1267 if (buf->ops) {
1268 buf->ops->release(pipe, buf);
1269 buf->ops = NULL;
1270 }
1271 }
Jens Axboeb92ce552006-04-11 13:52:07 +02001272
Jens Axboe9e971982008-01-29 21:05:57 +01001273 if (!bytes)
1274 bytes = ret;
Jens Axboeb92ce552006-04-11 13:52:07 +02001275
Jens Axboe9e971982008-01-29 21:05:57 +01001276 goto done;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001277}
1278EXPORT_SYMBOL(splice_direct_to_actor);
1279
1280static int direct_splice_actor(struct pipe_inode_info *pipe,
1281 struct splice_desc *sd)
1282{
Jens Axboe6a14b902007-06-14 13:08:55 +02001283 struct file *file = sd->u.file;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001284
Al Viro7995bd22013-06-20 18:58:36 +04001285 return do_splice_from(pipe, file, sd->opos, sd->total_len,
Changli Gao2cb4b052010-06-29 13:09:18 +02001286 sd->flags);
Jens Axboec66ab6f2007-06-12 21:17:17 +02001287}
1288
Jens Axboe932cc6d2007-06-21 13:10:21 +02001289/**
1290 * do_splice_direct - splices data directly between two files
1291 * @in: file to splice from
1292 * @ppos: input file offset
1293 * @out: file to splice to
Randy Dunlapacdb37c2013-06-22 19:44:08 -07001294 * @opos: output file offset
Jens Axboe932cc6d2007-06-21 13:10:21 +02001295 * @len: number of bytes to splice
1296 * @flags: splice modifier flags
1297 *
1298 * Description:
1299 * For use by do_sendfile(). splice can easily emulate sendfile, but
1300 * doing it in the application would incur an extra system call
1301 * (splice in + splice out, as compared to just sendfile()). So this helper
1302 * can splice directly through a process-private pipe.
1303 *
1304 */
Jens Axboec66ab6f2007-06-12 21:17:17 +02001305long do_splice_direct(struct file *in, loff_t *ppos, struct file *out,
Al Viro7995bd22013-06-20 18:58:36 +04001306 loff_t *opos, size_t len, unsigned int flags)
Jens Axboec66ab6f2007-06-12 21:17:17 +02001307{
1308 struct splice_desc sd = {
1309 .len = len,
1310 .total_len = len,
1311 .flags = flags,
1312 .pos = *ppos,
Jens Axboe6a14b902007-06-14 13:08:55 +02001313 .u.file = out,
Al Viro7995bd22013-06-20 18:58:36 +04001314 .opos = opos,
Jens Axboec66ab6f2007-06-12 21:17:17 +02001315 };
Jens Axboe51a92c02007-07-13 14:11:43 +02001316 long ret;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001317
Al Viro18c67cb2013-06-19 15:41:54 +04001318 if (unlikely(!(out->f_mode & FMODE_WRITE)))
1319 return -EBADF;
1320
1321 if (unlikely(out->f_flags & O_APPEND))
1322 return -EINVAL;
1323
1324 ret = rw_verify_area(WRITE, out, opos, len);
1325 if (unlikely(ret < 0))
1326 return ret;
1327
Jens Axboec66ab6f2007-06-12 21:17:17 +02001328 ret = splice_direct_to_actor(in, &sd, direct_splice_actor);
Jens Axboe51a92c02007-07-13 14:11:43 +02001329 if (ret > 0)
Tom Zanussia82c53a2008-05-09 13:28:36 +02001330 *ppos = sd.pos;
Jens Axboe51a92c02007-07-13 14:11:43 +02001331
Jens Axboec66ab6f2007-06-12 21:17:17 +02001332 return ret;
Jens Axboeb92ce552006-04-11 13:52:07 +02001333}
Miklos Szeredi1c118592014-10-24 00:14:35 +02001334EXPORT_SYMBOL(do_splice_direct);
Jens Axboeb92ce552006-04-11 13:52:07 +02001335
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001336static int splice_pipe_to_pipe(struct pipe_inode_info *ipipe,
1337 struct pipe_inode_info *opipe,
1338 size_t len, unsigned int flags);
Jens Axboeddac0d32006-11-04 12:49:32 +01001339
1340/*
Jens Axboe83f91352006-04-02 23:05:09 +02001341 * Determine where to splice to/from.
1342 */
Ingo Molnar529565d2006-04-10 15:18:58 +02001343static long do_splice(struct file *in, loff_t __user *off_in,
1344 struct file *out, loff_t __user *off_out,
1345 size_t len, unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +02001346{
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001347 struct pipe_inode_info *ipipe;
1348 struct pipe_inode_info *opipe;
Al Viro7995bd22013-06-20 18:58:36 +04001349 loff_t offset;
Jens Axboea4514eb2006-04-19 15:57:05 +02001350 long ret;
Jens Axboe5274f052006-03-30 15:15:30 +02001351
Linus Torvalds71993e62010-11-28 13:56:09 -08001352 ipipe = get_pipe_info(in);
1353 opipe = get_pipe_info(out);
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001354
1355 if (ipipe && opipe) {
1356 if (off_in || off_out)
1357 return -ESPIPE;
1358
1359 if (!(in->f_mode & FMODE_READ))
1360 return -EBADF;
1361
1362 if (!(out->f_mode & FMODE_WRITE))
1363 return -EBADF;
1364
1365 /* Splicing to self would be fun, but... */
1366 if (ipipe == opipe)
1367 return -EINVAL;
1368
1369 return splice_pipe_to_pipe(ipipe, opipe, len, flags);
1370 }
1371
1372 if (ipipe) {
Ingo Molnar529565d2006-04-10 15:18:58 +02001373 if (off_in)
1374 return -ESPIPE;
Jens Axboeb92ce552006-04-11 13:52:07 +02001375 if (off_out) {
Changli Gao19c9a492010-06-29 13:10:36 +02001376 if (!(out->f_mode & FMODE_PWRITE))
Jens Axboeb92ce552006-04-11 13:52:07 +02001377 return -EINVAL;
Jens Axboecbb7e572006-04-11 14:57:50 +02001378 if (copy_from_user(&offset, off_out, sizeof(loff_t)))
Jens Axboeb92ce552006-04-11 13:52:07 +02001379 return -EFAULT;
Al Viro7995bd22013-06-20 18:58:36 +04001380 } else {
1381 offset = out->f_pos;
1382 }
Ingo Molnar529565d2006-04-10 15:18:58 +02001383
Al Viro18c67cb2013-06-19 15:41:54 +04001384 if (unlikely(!(out->f_mode & FMODE_WRITE)))
1385 return -EBADF;
1386
1387 if (unlikely(out->f_flags & O_APPEND))
1388 return -EINVAL;
1389
1390 ret = rw_verify_area(WRITE, out, &offset, len);
1391 if (unlikely(ret < 0))
1392 return ret;
1393
Al Viro500368f2013-05-23 20:07:11 -04001394 file_start_write(out);
Al Viro7995bd22013-06-20 18:58:36 +04001395 ret = do_splice_from(ipipe, out, &offset, len, flags);
Al Viro500368f2013-05-23 20:07:11 -04001396 file_end_write(out);
Jens Axboea4514eb2006-04-19 15:57:05 +02001397
Al Viro7995bd22013-06-20 18:58:36 +04001398 if (!off_out)
1399 out->f_pos = offset;
1400 else if (copy_to_user(off_out, &offset, sizeof(loff_t)))
Jens Axboea4514eb2006-04-19 15:57:05 +02001401 ret = -EFAULT;
1402
1403 return ret;
Ingo Molnar529565d2006-04-10 15:18:58 +02001404 }
Jens Axboe5274f052006-03-30 15:15:30 +02001405
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001406 if (opipe) {
Ingo Molnar529565d2006-04-10 15:18:58 +02001407 if (off_out)
1408 return -ESPIPE;
Jens Axboeb92ce552006-04-11 13:52:07 +02001409 if (off_in) {
Changli Gao19c9a492010-06-29 13:10:36 +02001410 if (!(in->f_mode & FMODE_PREAD))
Jens Axboeb92ce552006-04-11 13:52:07 +02001411 return -EINVAL;
Jens Axboecbb7e572006-04-11 14:57:50 +02001412 if (copy_from_user(&offset, off_in, sizeof(loff_t)))
Jens Axboeb92ce552006-04-11 13:52:07 +02001413 return -EFAULT;
Al Viro7995bd22013-06-20 18:58:36 +04001414 } else {
1415 offset = in->f_pos;
1416 }
Ingo Molnar529565d2006-04-10 15:18:58 +02001417
Al Viro7995bd22013-06-20 18:58:36 +04001418 ret = do_splice_to(in, &offset, opipe, len, flags);
Jens Axboea4514eb2006-04-19 15:57:05 +02001419
Al Viro7995bd22013-06-20 18:58:36 +04001420 if (!off_in)
1421 in->f_pos = offset;
1422 else if (copy_to_user(off_in, &offset, sizeof(loff_t)))
Jens Axboea4514eb2006-04-19 15:57:05 +02001423 ret = -EFAULT;
1424
1425 return ret;
Ingo Molnar529565d2006-04-10 15:18:58 +02001426 }
Jens Axboe5274f052006-03-30 15:15:30 +02001427
1428 return -EINVAL;
1429}
1430
Jens Axboe912d35f2006-04-26 10:59:21 +02001431/*
1432 * Map an iov into an array of pages and offset/length tupples. With the
1433 * partial_page structure, we can map several non-contiguous ranges into
1434 * our ones pages[] map instead of splitting that operation into pieces.
1435 * Could easily be exported as a generic helper for other users, in which
1436 * case one would probably want to add a 'max_nr_pages' parameter as well.
1437 */
1438static int get_iovec_page_array(const struct iovec __user *iov,
1439 unsigned int nr_vecs, struct page **pages,
Eric Dumazetbd1a68b2012-04-04 11:08:51 +02001440 struct partial_page *partial, bool aligned,
Jens Axboe35f3d142010-05-20 10:43:18 +02001441 unsigned int pipe_buffers)
Jens Axboe912d35f2006-04-26 10:59:21 +02001442{
1443 int buffers = 0, error = 0;
1444
Jens Axboe912d35f2006-04-26 10:59:21 +02001445 while (nr_vecs) {
1446 unsigned long off, npages;
Linus Torvalds75723952007-10-01 13:17:28 -07001447 struct iovec entry;
Jens Axboe912d35f2006-04-26 10:59:21 +02001448 void __user *base;
1449 size_t len;
1450 int i;
1451
Linus Torvalds75723952007-10-01 13:17:28 -07001452 error = -EFAULT;
Nick Pigginbc40d732008-07-25 19:45:26 -07001453 if (copy_from_user(&entry, iov, sizeof(entry)))
Jens Axboe912d35f2006-04-26 10:59:21 +02001454 break;
Linus Torvalds75723952007-10-01 13:17:28 -07001455
1456 base = entry.iov_base;
1457 len = entry.iov_len;
Jens Axboe912d35f2006-04-26 10:59:21 +02001458
1459 /*
1460 * Sanity check this iovec. 0 read succeeds.
1461 */
Linus Torvalds75723952007-10-01 13:17:28 -07001462 error = 0;
Jens Axboe912d35f2006-04-26 10:59:21 +02001463 if (unlikely(!len))
1464 break;
1465 error = -EFAULT;
Bastian Blank712a30e2008-02-10 16:47:57 +02001466 if (!access_ok(VERIFY_READ, base, len))
Jens Axboe912d35f2006-04-26 10:59:21 +02001467 break;
1468
1469 /*
1470 * Get this base offset and number of pages, then map
1471 * in the user pages.
1472 */
1473 off = (unsigned long) base & ~PAGE_MASK;
Jens Axboe7afa6fd2006-05-01 20:02:33 +02001474
1475 /*
1476 * If asked for alignment, the offset must be zero and the
1477 * length a multiple of the PAGE_SIZE.
1478 */
1479 error = -EINVAL;
1480 if (aligned && (off || len & ~PAGE_MASK))
1481 break;
1482
Jens Axboe912d35f2006-04-26 10:59:21 +02001483 npages = (off + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
Jens Axboe35f3d142010-05-20 10:43:18 +02001484 if (npages > pipe_buffers - buffers)
1485 npages = pipe_buffers - buffers;
Jens Axboe912d35f2006-04-26 10:59:21 +02001486
Nick Pigginbc40d732008-07-25 19:45:26 -07001487 error = get_user_pages_fast((unsigned long)base, npages,
1488 0, &pages[buffers]);
Jens Axboe912d35f2006-04-26 10:59:21 +02001489
1490 if (unlikely(error <= 0))
1491 break;
1492
1493 /*
1494 * Fill this contiguous range into the partial page map.
1495 */
1496 for (i = 0; i < error; i++) {
Jens Axboe75914892006-05-02 12:57:18 +02001497 const int plen = min_t(size_t, len, PAGE_SIZE - off);
Jens Axboe912d35f2006-04-26 10:59:21 +02001498
1499 partial[buffers].offset = off;
1500 partial[buffers].len = plen;
1501
1502 off = 0;
1503 len -= plen;
1504 buffers++;
1505 }
1506
1507 /*
1508 * We didn't complete this iov, stop here since it probably
1509 * means we have to move some of this into a pipe to
1510 * be able to continue.
1511 */
1512 if (len)
1513 break;
1514
1515 /*
1516 * Don't continue if we mapped fewer pages than we asked for,
1517 * or if we mapped the max number of pages that we have
1518 * room for.
1519 */
Jens Axboe35f3d142010-05-20 10:43:18 +02001520 if (error < npages || buffers == pipe_buffers)
Jens Axboe912d35f2006-04-26 10:59:21 +02001521 break;
1522
1523 nr_vecs--;
1524 iov++;
1525 }
1526
Jens Axboe912d35f2006-04-26 10:59:21 +02001527 if (buffers)
1528 return buffers;
1529
1530 return error;
1531}
1532
Jens Axboe6a14b902007-06-14 13:08:55 +02001533static int pipe_to_user(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
1534 struct splice_desc *sd)
1535{
Al Viro6130f532014-02-03 18:19:51 -05001536 int n = copy_page_to_iter(buf->page, buf->offset, sd->len, sd->u.data);
1537 return n == sd->len ? n : -EFAULT;
Jens Axboe6a14b902007-06-14 13:08:55 +02001538}
1539
1540/*
1541 * For lack of a better implementation, implement vmsplice() to userspace
1542 * as a simple copy of the pipes pages to the user iov.
1543 */
Al Viro6130f532014-02-03 18:19:51 -05001544static long vmsplice_to_user(struct file *file, const struct iovec __user *uiov,
Jens Axboe6a14b902007-06-14 13:08:55 +02001545 unsigned long nr_segs, unsigned int flags)
1546{
1547 struct pipe_inode_info *pipe;
1548 struct splice_desc sd;
Jens Axboe6a14b902007-06-14 13:08:55 +02001549 long ret;
Al Viro6130f532014-02-03 18:19:51 -05001550 struct iovec iovstack[UIO_FASTIOV];
1551 struct iovec *iov = iovstack;
1552 struct iov_iter iter;
Jens Axboe6a14b902007-06-14 13:08:55 +02001553
Linus Torvalds71993e62010-11-28 13:56:09 -08001554 pipe = get_pipe_info(file);
Jens Axboe6a14b902007-06-14 13:08:55 +02001555 if (!pipe)
1556 return -EBADF;
1557
Al Viro345995f2015-03-21 19:17:55 -04001558 ret = import_iovec(READ, uiov, nr_segs,
1559 ARRAY_SIZE(iovstack), &iov, &iter);
1560 if (ret < 0)
1561 return ret;
Al Viro6130f532014-02-03 18:19:51 -05001562
Al Viro345995f2015-03-21 19:17:55 -04001563 sd.total_len = iov_iter_count(&iter);
Al Viro6130f532014-02-03 18:19:51 -05001564 sd.len = 0;
Al Viro6130f532014-02-03 18:19:51 -05001565 sd.flags = flags;
1566 sd.u.data = &iter;
1567 sd.pos = 0;
1568
Al Viro345995f2015-03-21 19:17:55 -04001569 if (sd.total_len) {
1570 pipe_lock(pipe);
1571 ret = __splice_from_pipe(pipe, &sd, pipe_to_user);
1572 pipe_unlock(pipe);
1573 }
Jens Axboe6a14b902007-06-14 13:08:55 +02001574
Al Viro345995f2015-03-21 19:17:55 -04001575 kfree(iov);
Jens Axboe6a14b902007-06-14 13:08:55 +02001576 return ret;
1577}
1578
Jens Axboe912d35f2006-04-26 10:59:21 +02001579/*
1580 * vmsplice splices a user address range into a pipe. It can be thought of
1581 * as splice-from-memory, where the regular splice is splice-from-file (or
1582 * to file). In both cases the output is a pipe, naturally.
Jens Axboe912d35f2006-04-26 10:59:21 +02001583 */
Jens Axboe6a14b902007-06-14 13:08:55 +02001584static long vmsplice_to_pipe(struct file *file, const struct iovec __user *iov,
1585 unsigned long nr_segs, unsigned int flags)
Jens Axboe912d35f2006-04-26 10:59:21 +02001586{
Jens Axboeddac0d32006-11-04 12:49:32 +01001587 struct pipe_inode_info *pipe;
Jens Axboe35f3d142010-05-20 10:43:18 +02001588 struct page *pages[PIPE_DEF_BUFFERS];
1589 struct partial_page partial[PIPE_DEF_BUFFERS];
Jens Axboe912d35f2006-04-26 10:59:21 +02001590 struct splice_pipe_desc spd = {
1591 .pages = pages,
1592 .partial = partial,
Eric Dumazet047fe362012-06-12 15:24:40 +02001593 .nr_pages_max = PIPE_DEF_BUFFERS,
Jens Axboe912d35f2006-04-26 10:59:21 +02001594 .flags = flags,
1595 .ops = &user_page_pipe_buf_ops,
Jens Axboebbdfc2f2007-11-06 23:29:47 -08001596 .spd_release = spd_release_page,
Jens Axboe912d35f2006-04-26 10:59:21 +02001597 };
Jens Axboe35f3d142010-05-20 10:43:18 +02001598 long ret;
Jens Axboe912d35f2006-04-26 10:59:21 +02001599
Linus Torvalds71993e62010-11-28 13:56:09 -08001600 pipe = get_pipe_info(file);
Jens Axboeddac0d32006-11-04 12:49:32 +01001601 if (!pipe)
Jens Axboe912d35f2006-04-26 10:59:21 +02001602 return -EBADF;
Jens Axboe912d35f2006-04-26 10:59:21 +02001603
Jens Axboe35f3d142010-05-20 10:43:18 +02001604 if (splice_grow_spd(pipe, &spd))
1605 return -ENOMEM;
Jens Axboe912d35f2006-04-26 10:59:21 +02001606
Jens Axboe35f3d142010-05-20 10:43:18 +02001607 spd.nr_pages = get_iovec_page_array(iov, nr_segs, spd.pages,
Eric Dumazetbd1a68b2012-04-04 11:08:51 +02001608 spd.partial, false,
Eric Dumazet047fe362012-06-12 15:24:40 +02001609 spd.nr_pages_max);
Jens Axboe35f3d142010-05-20 10:43:18 +02001610 if (spd.nr_pages <= 0)
1611 ret = spd.nr_pages;
1612 else
1613 ret = splice_to_pipe(pipe, &spd);
1614
Eric Dumazet047fe362012-06-12 15:24:40 +02001615 splice_shrink_spd(&spd);
Jens Axboe35f3d142010-05-20 10:43:18 +02001616 return ret;
Jens Axboe912d35f2006-04-26 10:59:21 +02001617}
1618
Jens Axboe6a14b902007-06-14 13:08:55 +02001619/*
1620 * Note that vmsplice only really supports true splicing _from_ user memory
1621 * to a pipe, not the other way around. Splicing from user memory is a simple
1622 * operation that can be supported without any funky alignment restrictions
1623 * or nasty vm tricks. We simply map in the user memory and fill them into
1624 * a pipe. The reverse isn't quite as easy, though. There are two possible
1625 * solutions for that:
1626 *
1627 * - memcpy() the data internally, at which point we might as well just
1628 * do a regular read() on the buffer anyway.
1629 * - Lots of nasty vm tricks, that are neither fast nor flexible (it
1630 * has restriction limitations on both ends of the pipe).
1631 *
1632 * Currently we punt and implement it as a normal copy, see pipe_to_user().
1633 *
1634 */
Heiko Carstens836f92a2009-01-14 14:14:33 +01001635SYSCALL_DEFINE4(vmsplice, int, fd, const struct iovec __user *, iov,
1636 unsigned long, nr_segs, unsigned int, flags)
Jens Axboe912d35f2006-04-26 10:59:21 +02001637{
Al Viro2903ff02012-08-28 12:52:22 -04001638 struct fd f;
Jens Axboe912d35f2006-04-26 10:59:21 +02001639 long error;
Jens Axboe912d35f2006-04-26 10:59:21 +02001640
Jens Axboe6a14b902007-06-14 13:08:55 +02001641 if (unlikely(nr_segs > UIO_MAXIOV))
1642 return -EINVAL;
1643 else if (unlikely(!nr_segs))
1644 return 0;
1645
Jens Axboe912d35f2006-04-26 10:59:21 +02001646 error = -EBADF;
Al Viro2903ff02012-08-28 12:52:22 -04001647 f = fdget(fd);
1648 if (f.file) {
1649 if (f.file->f_mode & FMODE_WRITE)
1650 error = vmsplice_to_pipe(f.file, iov, nr_segs, flags);
1651 else if (f.file->f_mode & FMODE_READ)
1652 error = vmsplice_to_user(f.file, iov, nr_segs, flags);
Jens Axboe912d35f2006-04-26 10:59:21 +02001653
Al Viro2903ff02012-08-28 12:52:22 -04001654 fdput(f);
Jens Axboe912d35f2006-04-26 10:59:21 +02001655 }
1656
1657 return error;
1658}
1659
Al Viro76b021d2013-03-02 10:19:56 -05001660#ifdef CONFIG_COMPAT
1661COMPAT_SYSCALL_DEFINE4(vmsplice, int, fd, const struct compat_iovec __user *, iov32,
1662 unsigned int, nr_segs, unsigned int, flags)
1663{
1664 unsigned i;
1665 struct iovec __user *iov;
1666 if (nr_segs > UIO_MAXIOV)
1667 return -EINVAL;
1668 iov = compat_alloc_user_space(nr_segs * sizeof(struct iovec));
1669 for (i = 0; i < nr_segs; i++) {
1670 struct compat_iovec v;
1671 if (get_user(v.iov_base, &iov32[i].iov_base) ||
1672 get_user(v.iov_len, &iov32[i].iov_len) ||
1673 put_user(compat_ptr(v.iov_base), &iov[i].iov_base) ||
1674 put_user(v.iov_len, &iov[i].iov_len))
1675 return -EFAULT;
1676 }
1677 return sys_vmsplice(fd, iov, nr_segs, flags);
1678}
1679#endif
1680
Heiko Carstens836f92a2009-01-14 14:14:33 +01001681SYSCALL_DEFINE6(splice, int, fd_in, loff_t __user *, off_in,
1682 int, fd_out, loff_t __user *, off_out,
1683 size_t, len, unsigned int, flags)
Jens Axboe5274f052006-03-30 15:15:30 +02001684{
Al Viro2903ff02012-08-28 12:52:22 -04001685 struct fd in, out;
Jens Axboe5274f052006-03-30 15:15:30 +02001686 long error;
Jens Axboe5274f052006-03-30 15:15:30 +02001687
1688 if (unlikely(!len))
1689 return 0;
1690
1691 error = -EBADF;
Al Viro2903ff02012-08-28 12:52:22 -04001692 in = fdget(fd_in);
1693 if (in.file) {
1694 if (in.file->f_mode & FMODE_READ) {
1695 out = fdget(fd_out);
1696 if (out.file) {
1697 if (out.file->f_mode & FMODE_WRITE)
1698 error = do_splice(in.file, off_in,
1699 out.file, off_out,
Ingo Molnar529565d2006-04-10 15:18:58 +02001700 len, flags);
Al Viro2903ff02012-08-28 12:52:22 -04001701 fdput(out);
Jens Axboe5274f052006-03-30 15:15:30 +02001702 }
1703 }
Al Viro2903ff02012-08-28 12:52:22 -04001704 fdput(in);
Jens Axboe5274f052006-03-30 15:15:30 +02001705 }
Jens Axboe5274f052006-03-30 15:15:30 +02001706 return error;
1707}
Jens Axboe70524492006-04-11 15:51:17 +02001708
1709/*
Jens Axboeaadd06e2006-07-10 11:00:01 +02001710 * Make sure there's data to read. Wait for input if we can, otherwise
1711 * return an appropriate error.
1712 */
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001713static int ipipe_prep(struct pipe_inode_info *pipe, unsigned int flags)
Jens Axboeaadd06e2006-07-10 11:00:01 +02001714{
1715 int ret;
1716
1717 /*
1718 * Check ->nrbufs without the inode lock first. This function
1719 * is speculative anyways, so missing one is ok.
1720 */
1721 if (pipe->nrbufs)
1722 return 0;
1723
1724 ret = 0;
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001725 pipe_lock(pipe);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001726
1727 while (!pipe->nrbufs) {
1728 if (signal_pending(current)) {
1729 ret = -ERESTARTSYS;
1730 break;
1731 }
1732 if (!pipe->writers)
1733 break;
1734 if (!pipe->waiting_writers) {
1735 if (flags & SPLICE_F_NONBLOCK) {
1736 ret = -EAGAIN;
1737 break;
1738 }
1739 }
1740 pipe_wait(pipe);
1741 }
1742
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001743 pipe_unlock(pipe);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001744 return ret;
1745}
1746
1747/*
1748 * Make sure there's writeable room. Wait for room if we can, otherwise
1749 * return an appropriate error.
1750 */
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001751static int opipe_prep(struct pipe_inode_info *pipe, unsigned int flags)
Jens Axboeaadd06e2006-07-10 11:00:01 +02001752{
1753 int ret;
1754
1755 /*
1756 * Check ->nrbufs without the inode lock first. This function
1757 * is speculative anyways, so missing one is ok.
1758 */
Jens Axboe35f3d142010-05-20 10:43:18 +02001759 if (pipe->nrbufs < pipe->buffers)
Jens Axboeaadd06e2006-07-10 11:00:01 +02001760 return 0;
1761
1762 ret = 0;
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001763 pipe_lock(pipe);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001764
Jens Axboe35f3d142010-05-20 10:43:18 +02001765 while (pipe->nrbufs >= pipe->buffers) {
Jens Axboeaadd06e2006-07-10 11:00:01 +02001766 if (!pipe->readers) {
1767 send_sig(SIGPIPE, current, 0);
1768 ret = -EPIPE;
1769 break;
1770 }
1771 if (flags & SPLICE_F_NONBLOCK) {
1772 ret = -EAGAIN;
1773 break;
1774 }
1775 if (signal_pending(current)) {
1776 ret = -ERESTARTSYS;
1777 break;
1778 }
1779 pipe->waiting_writers++;
1780 pipe_wait(pipe);
1781 pipe->waiting_writers--;
1782 }
1783
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001784 pipe_unlock(pipe);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001785 return ret;
1786}
1787
1788/*
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001789 * Splice contents of ipipe to opipe.
1790 */
1791static int splice_pipe_to_pipe(struct pipe_inode_info *ipipe,
1792 struct pipe_inode_info *opipe,
1793 size_t len, unsigned int flags)
1794{
1795 struct pipe_buffer *ibuf, *obuf;
1796 int ret = 0, nbuf;
1797 bool input_wakeup = false;
1798
1799
1800retry:
1801 ret = ipipe_prep(ipipe, flags);
1802 if (ret)
1803 return ret;
1804
1805 ret = opipe_prep(opipe, flags);
1806 if (ret)
1807 return ret;
1808
1809 /*
1810 * Potential ABBA deadlock, work around it by ordering lock
1811 * grabbing by pipe info address. Otherwise two different processes
1812 * could deadlock (one doing tee from A -> B, the other from B -> A).
1813 */
1814 pipe_double_lock(ipipe, opipe);
1815
1816 do {
1817 if (!opipe->readers) {
1818 send_sig(SIGPIPE, current, 0);
1819 if (!ret)
1820 ret = -EPIPE;
1821 break;
1822 }
1823
1824 if (!ipipe->nrbufs && !ipipe->writers)
1825 break;
1826
1827 /*
1828 * Cannot make any progress, because either the input
1829 * pipe is empty or the output pipe is full.
1830 */
Jens Axboe35f3d142010-05-20 10:43:18 +02001831 if (!ipipe->nrbufs || opipe->nrbufs >= opipe->buffers) {
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001832 /* Already processed some buffers, break */
1833 if (ret)
1834 break;
1835
1836 if (flags & SPLICE_F_NONBLOCK) {
1837 ret = -EAGAIN;
1838 break;
1839 }
1840
1841 /*
1842 * We raced with another reader/writer and haven't
1843 * managed to process any buffers. A zero return
1844 * value means EOF, so retry instead.
1845 */
1846 pipe_unlock(ipipe);
1847 pipe_unlock(opipe);
1848 goto retry;
1849 }
1850
1851 ibuf = ipipe->bufs + ipipe->curbuf;
Jens Axboe35f3d142010-05-20 10:43:18 +02001852 nbuf = (opipe->curbuf + opipe->nrbufs) & (opipe->buffers - 1);
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001853 obuf = opipe->bufs + nbuf;
1854
1855 if (len >= ibuf->len) {
1856 /*
1857 * Simply move the whole buffer from ipipe to opipe
1858 */
1859 *obuf = *ibuf;
1860 ibuf->ops = NULL;
1861 opipe->nrbufs++;
Jens Axboe35f3d142010-05-20 10:43:18 +02001862 ipipe->curbuf = (ipipe->curbuf + 1) & (ipipe->buffers - 1);
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001863 ipipe->nrbufs--;
1864 input_wakeup = true;
1865 } else {
1866 /*
1867 * Get a reference to this pipe buffer,
1868 * so we can copy the contents over.
1869 */
1870 ibuf->ops->get(ipipe, ibuf);
1871 *obuf = *ibuf;
1872
1873 /*
1874 * Don't inherit the gift flag, we need to
1875 * prevent multiple steals of this page.
1876 */
1877 obuf->flags &= ~PIPE_BUF_FLAG_GIFT;
1878
1879 obuf->len = len;
1880 opipe->nrbufs++;
1881 ibuf->offset += obuf->len;
1882 ibuf->len -= obuf->len;
1883 }
1884 ret += obuf->len;
1885 len -= obuf->len;
1886 } while (len);
1887
1888 pipe_unlock(ipipe);
1889 pipe_unlock(opipe);
1890
1891 /*
1892 * If we put data in the output pipe, wakeup any potential readers.
1893 */
Namhyung Kim825cdcb2011-05-23 19:58:53 +02001894 if (ret > 0)
1895 wakeup_pipe_readers(opipe);
1896
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001897 if (input_wakeup)
1898 wakeup_pipe_writers(ipipe);
1899
1900 return ret;
1901}
1902
1903/*
Jens Axboe70524492006-04-11 15:51:17 +02001904 * Link contents of ipipe to opipe.
1905 */
1906static int link_pipe(struct pipe_inode_info *ipipe,
1907 struct pipe_inode_info *opipe,
1908 size_t len, unsigned int flags)
1909{
1910 struct pipe_buffer *ibuf, *obuf;
Jens Axboeaadd06e2006-07-10 11:00:01 +02001911 int ret = 0, i = 0, nbuf;
Jens Axboe70524492006-04-11 15:51:17 +02001912
1913 /*
1914 * Potential ABBA deadlock, work around it by ordering lock
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001915 * grabbing by pipe info address. Otherwise two different processes
Jens Axboe70524492006-04-11 15:51:17 +02001916 * could deadlock (one doing tee from A -> B, the other from B -> A).
1917 */
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001918 pipe_double_lock(ipipe, opipe);
Jens Axboe70524492006-04-11 15:51:17 +02001919
Jens Axboeaadd06e2006-07-10 11:00:01 +02001920 do {
Jens Axboe70524492006-04-11 15:51:17 +02001921 if (!opipe->readers) {
1922 send_sig(SIGPIPE, current, 0);
1923 if (!ret)
1924 ret = -EPIPE;
1925 break;
1926 }
Jens Axboe70524492006-04-11 15:51:17 +02001927
1928 /*
Jens Axboeaadd06e2006-07-10 11:00:01 +02001929 * If we have iterated all input buffers or ran out of
1930 * output room, break.
Jens Axboe70524492006-04-11 15:51:17 +02001931 */
Jens Axboe35f3d142010-05-20 10:43:18 +02001932 if (i >= ipipe->nrbufs || opipe->nrbufs >= opipe->buffers)
Jens Axboe70524492006-04-11 15:51:17 +02001933 break;
Jens Axboeaadd06e2006-07-10 11:00:01 +02001934
Jens Axboe35f3d142010-05-20 10:43:18 +02001935 ibuf = ipipe->bufs + ((ipipe->curbuf + i) & (ipipe->buffers-1));
1936 nbuf = (opipe->curbuf + opipe->nrbufs) & (opipe->buffers - 1);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001937
Jens Axboe2a27250e2006-04-19 15:56:40 +02001938 /*
Jens Axboeaadd06e2006-07-10 11:00:01 +02001939 * Get a reference to this pipe buffer,
1940 * so we can copy the contents over.
Jens Axboe2a27250e2006-04-19 15:56:40 +02001941 */
Jens Axboeaadd06e2006-07-10 11:00:01 +02001942 ibuf->ops->get(ipipe, ibuf);
Jens Axboe70524492006-04-11 15:51:17 +02001943
Jens Axboeaadd06e2006-07-10 11:00:01 +02001944 obuf = opipe->bufs + nbuf;
1945 *obuf = *ibuf;
Jens Axboe70524492006-04-11 15:51:17 +02001946
Jens Axboeaadd06e2006-07-10 11:00:01 +02001947 /*
1948 * Don't inherit the gift flag, we need to
1949 * prevent multiple steals of this page.
1950 */
1951 obuf->flags &= ~PIPE_BUF_FLAG_GIFT;
1952
1953 if (obuf->len > len)
1954 obuf->len = len;
1955
1956 opipe->nrbufs++;
1957 ret += obuf->len;
1958 len -= obuf->len;
1959 i++;
1960 } while (len);
Jens Axboe70524492006-04-11 15:51:17 +02001961
Jens Axboe02cf01a2008-02-20 10:34:51 +01001962 /*
1963 * return EAGAIN if we have the potential of some data in the
1964 * future, otherwise just return 0
1965 */
1966 if (!ret && ipipe->waiting_writers && (flags & SPLICE_F_NONBLOCK))
1967 ret = -EAGAIN;
1968
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001969 pipe_unlock(ipipe);
1970 pipe_unlock(opipe);
Jens Axboe70524492006-04-11 15:51:17 +02001971
Jens Axboeaadd06e2006-07-10 11:00:01 +02001972 /*
1973 * If we put data in the output pipe, wakeup any potential readers.
1974 */
Namhyung Kim825cdcb2011-05-23 19:58:53 +02001975 if (ret > 0)
1976 wakeup_pipe_readers(opipe);
Jens Axboe70524492006-04-11 15:51:17 +02001977
1978 return ret;
1979}
1980
1981/*
1982 * This is a tee(1) implementation that works on pipes. It doesn't copy
1983 * any data, it simply references the 'in' pages on the 'out' pipe.
1984 * The 'flags' used are the SPLICE_F_* variants, currently the only
1985 * applicable one is SPLICE_F_NONBLOCK.
1986 */
1987static long do_tee(struct file *in, struct file *out, size_t len,
1988 unsigned int flags)
1989{
Linus Torvalds71993e62010-11-28 13:56:09 -08001990 struct pipe_inode_info *ipipe = get_pipe_info(in);
1991 struct pipe_inode_info *opipe = get_pipe_info(out);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001992 int ret = -EINVAL;
Jens Axboe70524492006-04-11 15:51:17 +02001993
1994 /*
Jens Axboeaadd06e2006-07-10 11:00:01 +02001995 * Duplicate the contents of ipipe to opipe without actually
1996 * copying the data.
Jens Axboe70524492006-04-11 15:51:17 +02001997 */
Jens Axboeaadd06e2006-07-10 11:00:01 +02001998 if (ipipe && opipe && ipipe != opipe) {
1999 /*
2000 * Keep going, unless we encounter an error. The ipipe/opipe
2001 * ordering doesn't really matter.
2002 */
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02002003 ret = ipipe_prep(ipipe, flags);
Jens Axboeaadd06e2006-07-10 11:00:01 +02002004 if (!ret) {
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02002005 ret = opipe_prep(opipe, flags);
Jens Axboe02cf01a2008-02-20 10:34:51 +01002006 if (!ret)
Jens Axboeaadd06e2006-07-10 11:00:01 +02002007 ret = link_pipe(ipipe, opipe, len, flags);
Jens Axboeaadd06e2006-07-10 11:00:01 +02002008 }
2009 }
Jens Axboe70524492006-04-11 15:51:17 +02002010
Jens Axboeaadd06e2006-07-10 11:00:01 +02002011 return ret;
Jens Axboe70524492006-04-11 15:51:17 +02002012}
2013
Heiko Carstens836f92a2009-01-14 14:14:33 +01002014SYSCALL_DEFINE4(tee, int, fdin, int, fdout, size_t, len, unsigned int, flags)
Jens Axboe70524492006-04-11 15:51:17 +02002015{
Al Viro2903ff02012-08-28 12:52:22 -04002016 struct fd in;
2017 int error;
Jens Axboe70524492006-04-11 15:51:17 +02002018
2019 if (unlikely(!len))
2020 return 0;
2021
2022 error = -EBADF;
Al Viro2903ff02012-08-28 12:52:22 -04002023 in = fdget(fdin);
2024 if (in.file) {
2025 if (in.file->f_mode & FMODE_READ) {
2026 struct fd out = fdget(fdout);
2027 if (out.file) {
2028 if (out.file->f_mode & FMODE_WRITE)
2029 error = do_tee(in.file, out.file,
2030 len, flags);
2031 fdput(out);
Jens Axboe70524492006-04-11 15:51:17 +02002032 }
2033 }
Al Viro2903ff02012-08-28 12:52:22 -04002034 fdput(in);
Jens Axboe70524492006-04-11 15:51:17 +02002035 }
2036
2037 return error;
2038}