blob: 13e5b4776e7aade28b0f2161b4eb9a27a8030f31 [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>
Jens Axboe5274f052006-03-30 15:15:30 +020034
Jens Axboe83f91352006-04-02 23:05:09 +020035/*
36 * Attempt to steal a page from a pipe buffer. This should perhaps go into
37 * a vm helper function, it's already simplified quite a bit by the
38 * addition of remove_mapping(). If success is returned, the caller may
39 * attempt to reuse this page for another destination.
40 */
Jens Axboe76ad4d12006-05-03 10:41:33 +020041static int page_cache_pipe_buf_steal(struct pipe_inode_info *pipe,
Jens Axboe5abc97a2006-03-30 15:16:46 +020042 struct pipe_buffer *buf)
43{
44 struct page *page = buf->page;
Jens Axboe9e94cd42006-06-20 15:01:12 +020045 struct address_space *mapping;
Jens Axboe5abc97a2006-03-30 15:16:46 +020046
Jens Axboe9e0267c2006-04-19 15:57:31 +020047 lock_page(page);
48
Jens Axboe9e94cd42006-06-20 15:01:12 +020049 mapping = page_mapping(page);
50 if (mapping) {
51 WARN_ON(!PageUptodate(page));
Jens Axboe5abc97a2006-03-30 15:16:46 +020052
Jens Axboe9e94cd42006-06-20 15:01:12 +020053 /*
54 * At least for ext2 with nobh option, we need to wait on
55 * writeback completing on this page, since we'll remove it
56 * from the pagecache. Otherwise truncate wont wait on the
57 * page, allowing the disk blocks to be reused by someone else
58 * before we actually wrote our data to them. fs corruption
59 * ensues.
60 */
61 wait_on_page_writeback(page);
Jens Axboead8d6f02006-04-02 23:10:32 +020062
David Howells266cf652009-04-03 16:42:36 +010063 if (page_has_private(page) &&
64 !try_to_release_page(page, GFP_KERNEL))
Jens Axboeca39d652008-05-20 21:27:41 +020065 goto out_unlock;
Jens Axboe4f6f0bd2006-04-02 23:04:46 +020066
Jens Axboe9e94cd42006-06-20 15:01:12 +020067 /*
68 * If we succeeded in removing the mapping, set LRU flag
69 * and return good.
70 */
71 if (remove_mapping(mapping, page)) {
72 buf->flags |= PIPE_BUF_FLAG_LRU;
73 return 0;
74 }
Jens Axboe9e0267c2006-04-19 15:57:31 +020075 }
Jens Axboe5abc97a2006-03-30 15:16:46 +020076
Jens Axboe9e94cd42006-06-20 15:01:12 +020077 /*
78 * Raced with truncate or failed to remove page from current
79 * address space, unlock and return failure.
80 */
Jens Axboeca39d652008-05-20 21:27:41 +020081out_unlock:
Jens Axboe9e94cd42006-06-20 15:01:12 +020082 unlock_page(page);
83 return 1;
Jens Axboe5abc97a2006-03-30 15:16:46 +020084}
85
Jens Axboe76ad4d12006-05-03 10:41:33 +020086static void page_cache_pipe_buf_release(struct pipe_inode_info *pipe,
Jens Axboe5274f052006-03-30 15:15:30 +020087 struct pipe_buffer *buf)
88{
89 page_cache_release(buf->page);
Jens Axboe14328732006-05-03 10:35:26 +020090 buf->flags &= ~PIPE_BUF_FLAG_LRU;
Jens Axboe5274f052006-03-30 15:15:30 +020091}
92
Jens Axboe08457182007-06-12 20:51:32 +020093/*
94 * Check whether the contents of buf is OK to access. Since the content
95 * is a page cache page, IO may be in flight.
96 */
Jens Axboecac36bb02007-06-14 13:10:48 +020097static int page_cache_pipe_buf_confirm(struct pipe_inode_info *pipe,
98 struct pipe_buffer *buf)
Jens Axboe5274f052006-03-30 15:15:30 +020099{
100 struct page *page = buf->page;
Jens Axboe49d0b212006-04-10 09:04:41 +0200101 int err;
Jens Axboe5274f052006-03-30 15:15:30 +0200102
103 if (!PageUptodate(page)) {
Jens Axboe49d0b212006-04-10 09:04:41 +0200104 lock_page(page);
105
106 /*
107 * Page got truncated/unhashed. This will cause a 0-byte
Ingo Molnar73d62d82006-04-11 13:57:21 +0200108 * splice, if this is the first page.
Jens Axboe49d0b212006-04-10 09:04:41 +0200109 */
110 if (!page->mapping) {
111 err = -ENODATA;
112 goto error;
113 }
114
115 /*
Ingo Molnar73d62d82006-04-11 13:57:21 +0200116 * Uh oh, read-error from disk.
Jens Axboe49d0b212006-04-10 09:04:41 +0200117 */
118 if (!PageUptodate(page)) {
119 err = -EIO;
120 goto error;
121 }
122
123 /*
Jens Axboef84d7512006-05-01 19:59:03 +0200124 * Page is ok afterall, we are done.
Jens Axboe49d0b212006-04-10 09:04:41 +0200125 */
Jens Axboe5274f052006-03-30 15:15:30 +0200126 unlock_page(page);
Jens Axboe5274f052006-03-30 15:15:30 +0200127 }
128
Jens Axboef84d7512006-05-01 19:59:03 +0200129 return 0;
Jens Axboe49d0b212006-04-10 09:04:41 +0200130error:
131 unlock_page(page);
Jens Axboef84d7512006-05-01 19:59:03 +0200132 return err;
Jens Axboe70524492006-04-11 15:51:17 +0200133}
134
Hugh Dickins708e3502011-07-25 17:12:32 -0700135const struct pipe_buf_operations page_cache_pipe_buf_ops = {
Jens Axboe5274f052006-03-30 15:15:30 +0200136 .can_merge = 0,
Jens Axboef84d7512006-05-01 19:59:03 +0200137 .map = generic_pipe_buf_map,
138 .unmap = generic_pipe_buf_unmap,
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 Axboef84d7512006-05-01 19:59:03 +0200157 .map = generic_pipe_buf_map,
158 .unmap = generic_pipe_buf_unmap,
Jens Axboecac36bb02007-06-14 13:10:48 +0200159 .confirm = generic_pipe_buf_confirm,
Jens Axboe912d35f2006-04-26 10:59:21 +0200160 .release = page_cache_pipe_buf_release,
161 .steal = user_page_pipe_buf_steal,
Jens Axboef84d7512006-05-01 19:59:03 +0200162 .get = generic_pipe_buf_get,
Jens Axboe912d35f2006-04-26 10:59:21 +0200163};
164
Namhyung Kim825cdcb2011-05-23 19:58:53 +0200165static void wakeup_pipe_readers(struct pipe_inode_info *pipe)
166{
167 smp_mb();
168 if (waitqueue_active(&pipe->wait))
169 wake_up_interruptible(&pipe->wait);
170 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
171}
172
Jens Axboe932cc6d2007-06-21 13:10:21 +0200173/**
174 * splice_to_pipe - fill passed data into a pipe
175 * @pipe: pipe to fill
176 * @spd: data to fill
177 *
178 * Description:
Randy Dunlap79685b82007-07-27 08:08:51 +0200179 * @spd contains a map of pages and len/offset tuples, along with
Jens Axboe932cc6d2007-06-21 13:10:21 +0200180 * the struct pipe_buf_operations associated with these pages. This
181 * function will link that data to the pipe.
182 *
Jens Axboe83f91352006-04-02 23:05:09 +0200183 */
Jens Axboed6b29d72007-06-04 09:59:47 +0200184ssize_t splice_to_pipe(struct pipe_inode_info *pipe,
185 struct splice_pipe_desc *spd)
Jens Axboe5274f052006-03-30 15:15:30 +0200186{
Jens Axboe00de00b2007-06-15 13:14:22 +0200187 unsigned int spd_pages = spd->nr_pages;
Jens Axboe912d35f2006-04-26 10:59:21 +0200188 int ret, do_wakeup, page_nr;
Jens Axboe5274f052006-03-30 15:15:30 +0200189
190 ret = 0;
191 do_wakeup = 0;
Jens Axboe912d35f2006-04-26 10:59:21 +0200192 page_nr = 0;
Jens Axboe5274f052006-03-30 15:15:30 +0200193
Miklos Szeredi61e0d472009-04-14 19:48:41 +0200194 pipe_lock(pipe);
Jens Axboe5274f052006-03-30 15:15:30 +0200195
Jens Axboe5274f052006-03-30 15:15:30 +0200196 for (;;) {
Ingo Molnar3a326a22006-04-10 15:18:35 +0200197 if (!pipe->readers) {
Jens Axboe5274f052006-03-30 15:15:30 +0200198 send_sig(SIGPIPE, current, 0);
199 if (!ret)
200 ret = -EPIPE;
201 break;
202 }
203
Jens Axboe35f3d142010-05-20 10:43:18 +0200204 if (pipe->nrbufs < pipe->buffers) {
205 int newbuf = (pipe->curbuf + pipe->nrbufs) & (pipe->buffers - 1);
Ingo Molnar3a326a22006-04-10 15:18:35 +0200206 struct pipe_buffer *buf = pipe->bufs + newbuf;
Jens Axboe5274f052006-03-30 15:15:30 +0200207
Jens Axboe912d35f2006-04-26 10:59:21 +0200208 buf->page = spd->pages[page_nr];
209 buf->offset = spd->partial[page_nr].offset;
210 buf->len = spd->partial[page_nr].len;
Jens Axboe497f9622007-06-11 12:00:45 +0200211 buf->private = spd->partial[page_nr].private;
Jens Axboe912d35f2006-04-26 10:59:21 +0200212 buf->ops = spd->ops;
Jens Axboe7afa6fd2006-05-01 20:02:33 +0200213 if (spd->flags & SPLICE_F_GIFT)
214 buf->flags |= PIPE_BUF_FLAG_GIFT;
215
Jens Axboe6f767b02006-04-11 13:53:56 +0200216 pipe->nrbufs++;
Jens Axboe912d35f2006-04-26 10:59:21 +0200217 page_nr++;
218 ret += buf->len;
219
Jens Axboe6f767b02006-04-11 13:53:56 +0200220 if (pipe->inode)
221 do_wakeup = 1;
Jens Axboe5274f052006-03-30 15:15:30 +0200222
Jens Axboe912d35f2006-04-26 10:59:21 +0200223 if (!--spd->nr_pages)
Jens Axboe5274f052006-03-30 15:15:30 +0200224 break;
Jens Axboe35f3d142010-05-20 10:43:18 +0200225 if (pipe->nrbufs < pipe->buffers)
Jens Axboe5274f052006-03-30 15:15:30 +0200226 continue;
227
228 break;
229 }
230
Jens Axboe912d35f2006-04-26 10:59:21 +0200231 if (spd->flags & SPLICE_F_NONBLOCK) {
Linus Torvalds29e35092006-04-02 12:46:35 -0700232 if (!ret)
233 ret = -EAGAIN;
234 break;
235 }
236
Jens Axboe5274f052006-03-30 15:15:30 +0200237 if (signal_pending(current)) {
238 if (!ret)
239 ret = -ERESTARTSYS;
240 break;
241 }
242
243 if (do_wakeup) {
Jens Axboec0bd1f62006-04-10 09:03:32 +0200244 smp_mb();
Ingo Molnar3a326a22006-04-10 15:18:35 +0200245 if (waitqueue_active(&pipe->wait))
246 wake_up_interruptible_sync(&pipe->wait);
247 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
Jens Axboe5274f052006-03-30 15:15:30 +0200248 do_wakeup = 0;
249 }
250
Ingo Molnar3a326a22006-04-10 15:18:35 +0200251 pipe->waiting_writers++;
252 pipe_wait(pipe);
253 pipe->waiting_writers--;
Jens Axboe5274f052006-03-30 15:15:30 +0200254 }
255
Miklos Szeredi61e0d472009-04-14 19:48:41 +0200256 pipe_unlock(pipe);
Jens Axboe5274f052006-03-30 15:15:30 +0200257
Namhyung Kim825cdcb2011-05-23 19:58:53 +0200258 if (do_wakeup)
259 wakeup_pipe_readers(pipe);
Jens Axboe5274f052006-03-30 15:15:30 +0200260
Jens Axboe00de00b2007-06-15 13:14:22 +0200261 while (page_nr < spd_pages)
Jens Axboebbdfc2f2007-11-06 23:29:47 -0800262 spd->spd_release(spd, page_nr++);
Jens Axboe5274f052006-03-30 15:15:30 +0200263
264 return ret;
265}
266
Hugh Dickins708e3502011-07-25 17:12:32 -0700267void spd_release_page(struct splice_pipe_desc *spd, unsigned int i)
Jens Axboebbdfc2f2007-11-06 23:29:47 -0800268{
269 page_cache_release(spd->pages[i]);
270}
271
Jens Axboe35f3d142010-05-20 10:43:18 +0200272/*
273 * Check if we need to grow the arrays holding pages and partial page
274 * descriptions.
275 */
Eric Dumazet047fe362012-06-12 15:24:40 +0200276int splice_grow_spd(const struct pipe_inode_info *pipe, struct splice_pipe_desc *spd)
Jens Axboe35f3d142010-05-20 10:43:18 +0200277{
Eric Dumazet047fe362012-06-12 15:24:40 +0200278 unsigned int buffers = ACCESS_ONCE(pipe->buffers);
279
280 spd->nr_pages_max = buffers;
281 if (buffers <= PIPE_DEF_BUFFERS)
Jens Axboe35f3d142010-05-20 10:43:18 +0200282 return 0;
283
Eric Dumazet047fe362012-06-12 15:24:40 +0200284 spd->pages = kmalloc(buffers * sizeof(struct page *), GFP_KERNEL);
285 spd->partial = kmalloc(buffers * sizeof(struct partial_page), GFP_KERNEL);
Jens Axboe35f3d142010-05-20 10:43:18 +0200286
287 if (spd->pages && spd->partial)
288 return 0;
289
290 kfree(spd->pages);
291 kfree(spd->partial);
292 return -ENOMEM;
293}
294
Eric Dumazet047fe362012-06-12 15:24:40 +0200295void splice_shrink_spd(struct splice_pipe_desc *spd)
Jens Axboe35f3d142010-05-20 10:43:18 +0200296{
Eric Dumazet047fe362012-06-12 15:24:40 +0200297 if (spd->nr_pages_max <= PIPE_DEF_BUFFERS)
Jens Axboe35f3d142010-05-20 10:43:18 +0200298 return;
299
300 kfree(spd->pages);
301 kfree(spd->partial);
302}
303
Ingo Molnar3a326a22006-04-10 15:18:35 +0200304static int
Jens Axboecbb7e572006-04-11 14:57:50 +0200305__generic_file_splice_read(struct file *in, loff_t *ppos,
306 struct pipe_inode_info *pipe, size_t len,
307 unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +0200308{
309 struct address_space *mapping = in->f_mapping;
Fengguang Wud8983912007-07-19 01:48:06 -0700310 unsigned int loff, nr_pages, req_pages;
Jens Axboe35f3d142010-05-20 10:43:18 +0200311 struct page *pages[PIPE_DEF_BUFFERS];
312 struct partial_page partial[PIPE_DEF_BUFFERS];
Jens Axboe5274f052006-03-30 15:15:30 +0200313 struct page *page;
Jens Axboe91ad66e2006-04-19 15:55:10 +0200314 pgoff_t index, end_index;
315 loff_t isize;
Jens Axboeeb207962006-04-27 11:05:22 +0200316 int error, page_nr;
Jens Axboe912d35f2006-04-26 10:59:21 +0200317 struct splice_pipe_desc spd = {
318 .pages = pages,
319 .partial = partial,
Eric Dumazet047fe362012-06-12 15:24:40 +0200320 .nr_pages_max = PIPE_DEF_BUFFERS,
Jens Axboe912d35f2006-04-26 10:59:21 +0200321 .flags = flags,
322 .ops = &page_cache_pipe_buf_ops,
Jens Axboebbdfc2f2007-11-06 23:29:47 -0800323 .spd_release = spd_release_page,
Jens Axboe912d35f2006-04-26 10:59:21 +0200324 };
Jens Axboe5274f052006-03-30 15:15:30 +0200325
Jens Axboe35f3d142010-05-20 10:43:18 +0200326 if (splice_grow_spd(pipe, &spd))
327 return -ENOMEM;
328
Jens Axboecbb7e572006-04-11 14:57:50 +0200329 index = *ppos >> PAGE_CACHE_SHIFT;
Jens Axboe912d35f2006-04-26 10:59:21 +0200330 loff = *ppos & ~PAGE_CACHE_MASK;
Fengguang Wud8983912007-07-19 01:48:06 -0700331 req_pages = (len + loff + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
Eric Dumazet047fe362012-06-12 15:24:40 +0200332 nr_pages = min(req_pages, spd.nr_pages_max);
Jens Axboe5274f052006-03-30 15:15:30 +0200333
334 /*
Jens Axboeeb207962006-04-27 11:05:22 +0200335 * Lookup the (hopefully) full range of pages we need.
336 */
Jens Axboe35f3d142010-05-20 10:43:18 +0200337 spd.nr_pages = find_get_pages_contig(mapping, index, nr_pages, spd.pages);
Fengguang Wu431a4822007-07-19 01:48:05 -0700338 index += spd.nr_pages;
Jens Axboeeb207962006-04-27 11:05:22 +0200339
340 /*
341 * If find_get_pages_contig() returned fewer pages than we needed,
Fengguang Wu431a4822007-07-19 01:48:05 -0700342 * readahead/allocate the rest and fill in the holes.
Jens Axboeeb207962006-04-27 11:05:22 +0200343 */
Fengguang Wu431a4822007-07-19 01:48:05 -0700344 if (spd.nr_pages < nr_pages)
Rusty Russellcf914a72007-07-19 01:48:08 -0700345 page_cache_sync_readahead(mapping, &in->f_ra, in,
346 index, req_pages - spd.nr_pages);
Fengguang Wu431a4822007-07-19 01:48:05 -0700347
Jens Axboe932cc6d2007-06-21 13:10:21 +0200348 error = 0;
Jens Axboeeb207962006-04-27 11:05:22 +0200349 while (spd.nr_pages < nr_pages) {
350 /*
351 * Page could be there, find_get_pages_contig() breaks on
352 * the first hole.
353 */
354 page = find_get_page(mapping, index);
355 if (!page) {
Jens Axboee27dedd2006-05-01 19:59:54 +0200356 /*
Jens Axboeeb207962006-04-27 11:05:22 +0200357 * page didn't exist, allocate one.
358 */
359 page = page_cache_alloc_cold(mapping);
360 if (!page)
361 break;
362
363 error = add_to_page_cache_lru(page, mapping, index,
Nick Piggin0ae0b5d2010-05-25 10:25:26 +0200364 GFP_KERNEL);
Jens Axboeeb207962006-04-27 11:05:22 +0200365 if (unlikely(error)) {
366 page_cache_release(page);
Jens Axboea0548872006-05-03 10:58:22 +0200367 if (error == -EEXIST)
368 continue;
Jens Axboeeb207962006-04-27 11:05:22 +0200369 break;
370 }
371 /*
372 * add_to_page_cache() locks the page, unlock it
373 * to avoid convoluting the logic below even more.
374 */
375 unlock_page(page);
376 }
377
Jens Axboe35f3d142010-05-20 10:43:18 +0200378 spd.pages[spd.nr_pages++] = page;
Jens Axboeeb207962006-04-27 11:05:22 +0200379 index++;
380 }
381
382 /*
383 * Now loop over the map and see if we need to start IO on any
384 * pages, fill in the partial map, etc.
385 */
386 index = *ppos >> PAGE_CACHE_SHIFT;
387 nr_pages = spd.nr_pages;
388 spd.nr_pages = 0;
389 for (page_nr = 0; page_nr < nr_pages; page_nr++) {
Jens Axboe82aa5d62006-04-20 13:05:48 +0200390 unsigned int this_len;
391
392 if (!len)
393 break;
394
395 /*
396 * this_len is the max we'll use from this page
397 */
Andrew Mortonba5f5d92006-04-25 15:33:34 +0200398 this_len = min_t(unsigned long, len, PAGE_CACHE_SIZE - loff);
Jens Axboe35f3d142010-05-20 10:43:18 +0200399 page = spd.pages[page_nr];
Jens Axboe7480a902006-04-11 13:52:47 +0200400
Fengguang Wua08a1662007-07-19 01:48:03 -0700401 if (PageReadahead(page))
Rusty Russellcf914a72007-07-19 01:48:08 -0700402 page_cache_async_readahead(mapping, &in->f_ra, in,
Fengguang Wud8983912007-07-19 01:48:06 -0700403 page, index, req_pages - page_nr);
Fengguang Wua08a1662007-07-19 01:48:03 -0700404
Jens Axboe7480a902006-04-11 13:52:47 +0200405 /*
406 * If the page isn't uptodate, we may need to start io on it
407 */
408 if (!PageUptodate(page)) {
Miklos Szeredi69650312010-08-03 12:48:50 +0200409 lock_page(page);
Jens Axboe7480a902006-04-11 13:52:47 +0200410
411 /*
Miklos Szeredi32502b82008-07-04 09:35:17 +0200412 * Page was truncated, or invalidated by the
413 * filesystem. Redo the find/create, but this time the
414 * page is kept locked, so there's no chance of another
415 * race with truncate/invalidate.
Jens Axboe7480a902006-04-11 13:52:47 +0200416 */
417 if (!page->mapping) {
418 unlock_page(page);
Miklos Szeredi32502b82008-07-04 09:35:17 +0200419 page = find_or_create_page(mapping, index,
420 mapping_gfp_mask(mapping));
421
422 if (!page) {
423 error = -ENOMEM;
424 break;
425 }
Jens Axboe35f3d142010-05-20 10:43:18 +0200426 page_cache_release(spd.pages[page_nr]);
427 spd.pages[page_nr] = page;
Jens Axboe7480a902006-04-11 13:52:47 +0200428 }
429 /*
430 * page was already under io and is now done, great
431 */
432 if (PageUptodate(page)) {
433 unlock_page(page);
434 goto fill_it;
435 }
436
Jens Axboe7480a902006-04-11 13:52:47 +0200437 /*
438 * need to read in the page
439 */
440 error = mapping->a_ops->readpage(in, page);
Jens Axboe7480a902006-04-11 13:52:47 +0200441 if (unlikely(error)) {
Jens Axboeeb207962006-04-27 11:05:22 +0200442 /*
443 * We really should re-lookup the page here,
444 * but it complicates things a lot. Instead
445 * lets just do what we already stored, and
446 * we'll get it the next time we are called.
447 */
Jens Axboe7480a902006-04-11 13:52:47 +0200448 if (error == AOP_TRUNCATED_PAGE)
Jens Axboeeb207962006-04-27 11:05:22 +0200449 error = 0;
450
Jens Axboe7480a902006-04-11 13:52:47 +0200451 break;
452 }
Jens Axboe620a3242007-06-07 09:39:42 +0200453 }
454fill_it:
455 /*
456 * i_size must be checked after PageUptodate.
457 */
458 isize = i_size_read(mapping->host);
459 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
460 if (unlikely(!isize || index > end_index))
461 break;
462
463 /*
464 * if this is the last page, see if we need to shrink
465 * the length and stop
466 */
467 if (end_index == index) {
468 unsigned int plen;
Jens Axboe91ad66e2006-04-19 15:55:10 +0200469
470 /*
Jens Axboe620a3242007-06-07 09:39:42 +0200471 * max good bytes in this page
Jens Axboe91ad66e2006-04-19 15:55:10 +0200472 */
Jens Axboe620a3242007-06-07 09:39:42 +0200473 plen = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
474 if (plen <= loff)
Jens Axboe91ad66e2006-04-19 15:55:10 +0200475 break;
Jens Axboe91ad66e2006-04-19 15:55:10 +0200476
477 /*
Jens Axboe620a3242007-06-07 09:39:42 +0200478 * force quit after adding this page
Jens Axboe91ad66e2006-04-19 15:55:10 +0200479 */
Jens Axboe620a3242007-06-07 09:39:42 +0200480 this_len = min(this_len, plen - loff);
481 len = this_len;
Jens Axboe7480a902006-04-11 13:52:47 +0200482 }
Jens Axboe620a3242007-06-07 09:39:42 +0200483
Jens Axboe35f3d142010-05-20 10:43:18 +0200484 spd.partial[page_nr].offset = loff;
485 spd.partial[page_nr].len = this_len;
Jens Axboe82aa5d62006-04-20 13:05:48 +0200486 len -= this_len;
Jens Axboe91ad66e2006-04-19 15:55:10 +0200487 loff = 0;
Jens Axboeeb207962006-04-27 11:05:22 +0200488 spd.nr_pages++;
489 index++;
Jens Axboe5274f052006-03-30 15:15:30 +0200490 }
491
Jens Axboeeb207962006-04-27 11:05:22 +0200492 /*
Hugh Dickins475ecad2007-06-07 09:36:00 +0200493 * Release any pages at the end, if we quit early. 'page_nr' is how far
Jens Axboeeb207962006-04-27 11:05:22 +0200494 * we got, 'nr_pages' is how many pages are in the map.
495 */
496 while (page_nr < nr_pages)
Jens Axboe35f3d142010-05-20 10:43:18 +0200497 page_cache_release(spd.pages[page_nr++]);
Fengguang Wuf4e6b492007-10-16 01:24:33 -0700498 in->f_ra.prev_pos = (loff_t)index << PAGE_CACHE_SHIFT;
Jens Axboeeb207962006-04-27 11:05:22 +0200499
Jens Axboe912d35f2006-04-26 10:59:21 +0200500 if (spd.nr_pages)
Jens Axboe35f3d142010-05-20 10:43:18 +0200501 error = splice_to_pipe(pipe, &spd);
Jens Axboe5274f052006-03-30 15:15:30 +0200502
Eric Dumazet047fe362012-06-12 15:24:40 +0200503 splice_shrink_spd(&spd);
Jens Axboe7480a902006-04-11 13:52:47 +0200504 return error;
Jens Axboe5274f052006-03-30 15:15:30 +0200505}
506
Jens Axboe83f91352006-04-02 23:05:09 +0200507/**
508 * generic_file_splice_read - splice data from file to a pipe
509 * @in: file to splice from
Jens Axboe932cc6d2007-06-21 13:10:21 +0200510 * @ppos: position in @in
Jens Axboe83f91352006-04-02 23:05:09 +0200511 * @pipe: pipe to splice to
512 * @len: number of bytes to splice
513 * @flags: splice modifier flags
514 *
Jens Axboe932cc6d2007-06-21 13:10:21 +0200515 * Description:
516 * Will read pages from given file and fill them into a pipe. Can be
517 * used as long as the address_space operations for the source implements
518 * a readpage() hook.
519 *
Jens Axboe83f91352006-04-02 23:05:09 +0200520 */
Jens Axboecbb7e572006-04-11 14:57:50 +0200521ssize_t generic_file_splice_read(struct file *in, loff_t *ppos,
522 struct pipe_inode_info *pipe, size_t len,
523 unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +0200524{
Jens Axboed366d3982007-06-01 14:54:11 +0200525 loff_t isize, left;
Jens Axboe8191ecd2008-04-10 08:24:25 +0200526 int ret;
Jens Axboed366d3982007-06-01 14:54:11 +0200527
528 isize = i_size_read(in->f_mapping->host);
529 if (unlikely(*ppos >= isize))
530 return 0;
531
532 left = isize - *ppos;
533 if (unlikely(left < len))
534 len = left;
Jens Axboe5274f052006-03-30 15:15:30 +0200535
Jens Axboe8191ecd2008-04-10 08:24:25 +0200536 ret = __generic_file_splice_read(in, ppos, pipe, len, flags);
Miklos Szeredi723590e2009-08-15 08:43:22 +0200537 if (ret > 0) {
Jens Axboecbb7e572006-04-11 14:57:50 +0200538 *ppos += ret;
Miklos Szeredi723590e2009-08-15 08:43:22 +0200539 file_accessed(in);
540 }
Jens Axboe5274f052006-03-30 15:15:30 +0200541
542 return ret;
543}
Jens Axboe059a8f32006-04-02 23:06:05 +0200544EXPORT_SYMBOL(generic_file_splice_read);
545
Miklos Szeredi68181732009-05-07 15:37:36 +0200546static const struct pipe_buf_operations default_pipe_buf_ops = {
547 .can_merge = 0,
548 .map = generic_pipe_buf_map,
549 .unmap = generic_pipe_buf_unmap,
550 .confirm = generic_pipe_buf_confirm,
551 .release = generic_pipe_buf_release,
552 .steal = generic_pipe_buf_steal,
553 .get = generic_pipe_buf_get,
554};
555
556static ssize_t kernel_readv(struct file *file, const struct iovec *vec,
557 unsigned long vlen, loff_t offset)
558{
559 mm_segment_t old_fs;
560 loff_t pos = offset;
561 ssize_t res;
562
563 old_fs = get_fs();
564 set_fs(get_ds());
565 /* The cast to a user pointer is valid due to the set_fs() */
566 res = vfs_readv(file, (const struct iovec __user *)vec, vlen, &pos);
567 set_fs(old_fs);
568
569 return res;
570}
571
Miklos Szeredib2858d72009-05-19 11:37:46 +0200572static ssize_t kernel_write(struct file *file, const char *buf, size_t count,
573 loff_t pos)
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +0200574{
575 mm_segment_t old_fs;
576 ssize_t res;
577
578 old_fs = get_fs();
579 set_fs(get_ds());
580 /* The cast to a user pointer is valid due to the set_fs() */
Miklos Szeredib2858d72009-05-19 11:37:46 +0200581 res = vfs_write(file, (const char __user *)buf, count, &pos);
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +0200582 set_fs(old_fs);
583
584 return res;
585}
586
Miklos Szeredi68181732009-05-07 15:37:36 +0200587ssize_t default_file_splice_read(struct file *in, loff_t *ppos,
588 struct pipe_inode_info *pipe, size_t len,
589 unsigned int flags)
590{
591 unsigned int nr_pages;
592 unsigned int nr_freed;
593 size_t offset;
Jens Axboe35f3d142010-05-20 10:43:18 +0200594 struct page *pages[PIPE_DEF_BUFFERS];
595 struct partial_page partial[PIPE_DEF_BUFFERS];
596 struct iovec *vec, __vec[PIPE_DEF_BUFFERS];
Miklos Szeredi68181732009-05-07 15:37:36 +0200597 ssize_t res;
598 size_t this_len;
599 int error;
600 int i;
601 struct splice_pipe_desc spd = {
602 .pages = pages,
603 .partial = partial,
Eric Dumazet047fe362012-06-12 15:24:40 +0200604 .nr_pages_max = PIPE_DEF_BUFFERS,
Miklos Szeredi68181732009-05-07 15:37:36 +0200605 .flags = flags,
606 .ops = &default_pipe_buf_ops,
607 .spd_release = spd_release_page,
608 };
609
Jens Axboe35f3d142010-05-20 10:43:18 +0200610 if (splice_grow_spd(pipe, &spd))
611 return -ENOMEM;
612
613 res = -ENOMEM;
614 vec = __vec;
Eric Dumazet047fe362012-06-12 15:24:40 +0200615 if (spd.nr_pages_max > PIPE_DEF_BUFFERS) {
616 vec = kmalloc(spd.nr_pages_max * sizeof(struct iovec), GFP_KERNEL);
Jens Axboe35f3d142010-05-20 10:43:18 +0200617 if (!vec)
618 goto shrink_ret;
619 }
620
Miklos Szeredi68181732009-05-07 15:37:36 +0200621 offset = *ppos & ~PAGE_CACHE_MASK;
622 nr_pages = (len + offset + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
623
Eric Dumazet047fe362012-06-12 15:24:40 +0200624 for (i = 0; i < nr_pages && i < spd.nr_pages_max && len; i++) {
Miklos Szeredi68181732009-05-07 15:37:36 +0200625 struct page *page;
626
Jens Axboe4f231222009-05-13 08:35:35 +0200627 page = alloc_page(GFP_USER);
Miklos Szeredi68181732009-05-07 15:37:36 +0200628 error = -ENOMEM;
629 if (!page)
630 goto err;
631
632 this_len = min_t(size_t, len, PAGE_CACHE_SIZE - offset);
Jens Axboe4f231222009-05-13 08:35:35 +0200633 vec[i].iov_base = (void __user *) page_address(page);
Miklos Szeredi68181732009-05-07 15:37:36 +0200634 vec[i].iov_len = this_len;
Jens Axboe35f3d142010-05-20 10:43:18 +0200635 spd.pages[i] = page;
Miklos Szeredi68181732009-05-07 15:37:36 +0200636 spd.nr_pages++;
637 len -= this_len;
638 offset = 0;
639 }
640
641 res = kernel_readv(in, vec, spd.nr_pages, *ppos);
Andrew Morton77f6bf52009-05-14 09:49:44 +0200642 if (res < 0) {
643 error = res;
Miklos Szeredi68181732009-05-07 15:37:36 +0200644 goto err;
Andrew Morton77f6bf52009-05-14 09:49:44 +0200645 }
Miklos Szeredi68181732009-05-07 15:37:36 +0200646
647 error = 0;
648 if (!res)
649 goto err;
650
651 nr_freed = 0;
652 for (i = 0; i < spd.nr_pages; i++) {
Miklos Szeredi68181732009-05-07 15:37:36 +0200653 this_len = min_t(size_t, vec[i].iov_len, res);
Jens Axboe35f3d142010-05-20 10:43:18 +0200654 spd.partial[i].offset = 0;
655 spd.partial[i].len = this_len;
Miklos Szeredi68181732009-05-07 15:37:36 +0200656 if (!this_len) {
Jens Axboe35f3d142010-05-20 10:43:18 +0200657 __free_page(spd.pages[i]);
658 spd.pages[i] = NULL;
Miklos Szeredi68181732009-05-07 15:37:36 +0200659 nr_freed++;
660 }
661 res -= this_len;
662 }
663 spd.nr_pages -= nr_freed;
664
665 res = splice_to_pipe(pipe, &spd);
666 if (res > 0)
667 *ppos += res;
668
Jens Axboe35f3d142010-05-20 10:43:18 +0200669shrink_ret:
670 if (vec != __vec)
671 kfree(vec);
Eric Dumazet047fe362012-06-12 15:24:40 +0200672 splice_shrink_spd(&spd);
Miklos Szeredi68181732009-05-07 15:37:36 +0200673 return res;
674
675err:
Jens Axboe4f231222009-05-13 08:35:35 +0200676 for (i = 0; i < spd.nr_pages; i++)
Jens Axboe35f3d142010-05-20 10:43:18 +0200677 __free_page(spd.pages[i]);
Jens Axboe4f231222009-05-13 08:35:35 +0200678
Jens Axboe35f3d142010-05-20 10:43:18 +0200679 res = error;
680 goto shrink_ret;
Miklos Szeredi68181732009-05-07 15:37:36 +0200681}
682EXPORT_SYMBOL(default_file_splice_read);
683
Jens Axboe5274f052006-03-30 15:15:30 +0200684/*
Jens Axboe4f6f0bd2006-04-02 23:04:46 +0200685 * Send 'sd->len' bytes to socket from 'sd->file' at position 'sd->pos'
Jens Axboe016b6612006-04-25 15:42:00 +0200686 * using sendpage(). Return the number of bytes sent.
Jens Axboe5274f052006-03-30 15:15:30 +0200687 */
Jens Axboe76ad4d12006-05-03 10:41:33 +0200688static int pipe_to_sendpage(struct pipe_inode_info *pipe,
Jens Axboe5274f052006-03-30 15:15:30 +0200689 struct pipe_buffer *buf, struct splice_desc *sd)
690{
Jens Axboe6a14b902007-06-14 13:08:55 +0200691 struct file *file = sd->u.file;
Jens Axboe5274f052006-03-30 15:15:30 +0200692 loff_t pos = sd->pos;
Michał Mirosława8adbe32010-12-17 08:56:44 +0100693 int more;
Jens Axboe5274f052006-03-30 15:15:30 +0200694
Michał Mirosława8adbe32010-12-17 08:56:44 +0100695 if (!likely(file->f_op && file->f_op->sendpage))
696 return -EINVAL;
Jens Axboe5274f052006-03-30 15:15:30 +0200697
Eric Dumazet35f9c092012-04-05 03:05:35 +0000698 more = (sd->flags & SPLICE_F_MORE) ? MSG_MORE : 0;
699 if (sd->len < sd->total_len)
700 more |= MSG_SENDPAGE_NOTLAST;
Michał Mirosława8adbe32010-12-17 08:56:44 +0100701 return file->f_op->sendpage(file, buf->page, buf->offset,
702 sd->len, &pos, more);
Jens Axboe5274f052006-03-30 15:15:30 +0200703}
704
705/*
706 * This is a little more tricky than the file -> pipe splicing. There are
707 * basically three cases:
708 *
709 * - Destination page already exists in the address space and there
710 * are users of it. For that case we have no other option that
711 * copying the data. Tough luck.
712 * - Destination page already exists in the address space, but there
713 * are no users of it. Make sure it's uptodate, then drop it. Fall
714 * through to last case.
715 * - Destination page does not exist, we can add the pipe page to
716 * the page cache and avoid the copy.
717 *
Jens Axboe83f91352006-04-02 23:05:09 +0200718 * If asked to move pages to the output file (SPLICE_F_MOVE is set in
719 * sd->flags), we attempt to migrate pages from the pipe to the output
720 * file address space page cache. This is possible if no one else has
721 * the pipe page referenced outside of the pipe and page cache. If
722 * SPLICE_F_MOVE isn't set, or we cannot move the page, we simply create
723 * a new page in the output file page cache and fill/dirty that.
Jens Axboe5274f052006-03-30 15:15:30 +0200724 */
Miklos Szeredi328eaab2009-04-14 19:48:39 +0200725int pipe_to_file(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
726 struct splice_desc *sd)
Jens Axboe5274f052006-03-30 15:15:30 +0200727{
Jens Axboe6a14b902007-06-14 13:08:55 +0200728 struct file *file = sd->u.file;
Jens Axboe5274f052006-03-30 15:15:30 +0200729 struct address_space *mapping = file->f_mapping;
Jens Axboe016b6612006-04-25 15:42:00 +0200730 unsigned int offset, this_len;
Jens Axboe5274f052006-03-30 15:15:30 +0200731 struct page *page;
Nick Pigginafddba42007-10-16 01:25:01 -0700732 void *fsdata;
Jens Axboe3e7ee3e2006-04-02 23:11:04 +0200733 int ret;
Jens Axboe5274f052006-03-30 15:15:30 +0200734
Jens Axboe5274f052006-03-30 15:15:30 +0200735 offset = sd->pos & ~PAGE_CACHE_MASK;
736
Jens Axboe016b6612006-04-25 15:42:00 +0200737 this_len = sd->len;
738 if (this_len + offset > PAGE_CACHE_SIZE)
739 this_len = PAGE_CACHE_SIZE - offset;
740
Nick Pigginafddba42007-10-16 01:25:01 -0700741 ret = pagecache_write_begin(file, mapping, sd->pos, this_len,
742 AOP_FLAG_UNINTERRUPTIBLE, &page, &fsdata);
743 if (unlikely(ret))
744 goto out;
Jens Axboe5274f052006-03-30 15:15:30 +0200745
Jens Axboe0568b402006-05-01 19:50:48 +0200746 if (buf->page != page) {
Jens Axboe76ad4d12006-05-03 10:41:33 +0200747 char *src = buf->ops->map(pipe, buf, 1);
Cong Wange8e3c3d2011-11-25 23:14:27 +0800748 char *dst = kmap_atomic(page);
Jens Axboe5abc97a2006-03-30 15:16:46 +0200749
Jens Axboe016b6612006-04-25 15:42:00 +0200750 memcpy(dst + offset, src + buf->offset, this_len);
Jens Axboe5abc97a2006-03-30 15:16:46 +0200751 flush_dcache_page(page);
Cong Wange8e3c3d2011-11-25 23:14:27 +0800752 kunmap_atomic(dst);
Jens Axboe76ad4d12006-05-03 10:41:33 +0200753 buf->ops->unmap(pipe, buf, src);
Jens Axboe5abc97a2006-03-30 15:16:46 +0200754 }
Nick Pigginafddba42007-10-16 01:25:01 -0700755 ret = pagecache_write_end(file, mapping, sd->pos, this_len, this_len,
756 page, fsdata);
Jens Axboe5274f052006-03-30 15:15:30 +0200757out:
Jens Axboe5274f052006-03-30 15:15:30 +0200758 return ret;
759}
Miklos Szeredi328eaab2009-04-14 19:48:39 +0200760EXPORT_SYMBOL(pipe_to_file);
Jens Axboe5274f052006-03-30 15:15:30 +0200761
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200762static void wakeup_pipe_writers(struct pipe_inode_info *pipe)
763{
764 smp_mb();
765 if (waitqueue_active(&pipe->wait))
766 wake_up_interruptible(&pipe->wait);
767 kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
768}
769
770/**
771 * splice_from_pipe_feed - feed available data from a pipe to a file
772 * @pipe: pipe to splice from
773 * @sd: information to @actor
774 * @actor: handler that splices the data
775 *
776 * Description:
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200777 * This function loops over the pipe and calls @actor to do the
778 * actual moving of a single struct pipe_buffer to the desired
779 * destination. It returns when there's no more buffers left in
780 * the pipe or if the requested number of bytes (@sd->total_len)
781 * have been copied. It returns a positive number (one) if the
782 * pipe needs to be filled with more data, zero if the required
783 * number of bytes have been copied and -errno on error.
784 *
785 * This, together with splice_from_pipe_{begin,end,next}, may be
786 * used to implement the functionality of __splice_from_pipe() when
787 * locking is required around copying the pipe buffers to the
788 * destination.
789 */
790int splice_from_pipe_feed(struct pipe_inode_info *pipe, struct splice_desc *sd,
791 splice_actor *actor)
792{
793 int ret;
794
795 while (pipe->nrbufs) {
796 struct pipe_buffer *buf = pipe->bufs + pipe->curbuf;
797 const struct pipe_buf_operations *ops = buf->ops;
798
799 sd->len = buf->len;
800 if (sd->len > sd->total_len)
801 sd->len = sd->total_len;
802
Michał Mirosława8adbe32010-12-17 08:56:44 +0100803 ret = buf->ops->confirm(pipe, buf);
804 if (unlikely(ret)) {
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200805 if (ret == -ENODATA)
806 ret = 0;
807 return ret;
808 }
Michał Mirosława8adbe32010-12-17 08:56:44 +0100809
810 ret = actor(pipe, buf, sd);
811 if (ret <= 0)
812 return ret;
813
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200814 buf->offset += ret;
815 buf->len -= ret;
816
817 sd->num_spliced += ret;
818 sd->len -= ret;
819 sd->pos += ret;
820 sd->total_len -= ret;
821
822 if (!buf->len) {
823 buf->ops = NULL;
824 ops->release(pipe, buf);
Jens Axboe35f3d142010-05-20 10:43:18 +0200825 pipe->curbuf = (pipe->curbuf + 1) & (pipe->buffers - 1);
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200826 pipe->nrbufs--;
827 if (pipe->inode)
828 sd->need_wakeup = true;
829 }
830
831 if (!sd->total_len)
832 return 0;
833 }
834
835 return 1;
836}
837EXPORT_SYMBOL(splice_from_pipe_feed);
838
839/**
840 * splice_from_pipe_next - wait for some data to splice from
841 * @pipe: pipe to splice from
842 * @sd: information about the splice operation
843 *
844 * Description:
845 * This function will wait for some data and return a positive
846 * value (one) if pipe buffers are available. It will return zero
847 * or -errno if no more data needs to be spliced.
848 */
849int splice_from_pipe_next(struct pipe_inode_info *pipe, struct splice_desc *sd)
850{
851 while (!pipe->nrbufs) {
852 if (!pipe->writers)
853 return 0;
854
855 if (!pipe->waiting_writers && sd->num_spliced)
856 return 0;
857
858 if (sd->flags & SPLICE_F_NONBLOCK)
859 return -EAGAIN;
860
861 if (signal_pending(current))
862 return -ERESTARTSYS;
863
864 if (sd->need_wakeup) {
865 wakeup_pipe_writers(pipe);
866 sd->need_wakeup = false;
867 }
868
869 pipe_wait(pipe);
870 }
871
872 return 1;
873}
874EXPORT_SYMBOL(splice_from_pipe_next);
875
876/**
877 * splice_from_pipe_begin - start splicing from pipe
Randy Dunlapb80901b2009-04-16 19:09:55 -0700878 * @sd: information about the splice operation
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200879 *
880 * Description:
881 * This function should be called before a loop containing
882 * splice_from_pipe_next() and splice_from_pipe_feed() to
883 * initialize the necessary fields of @sd.
884 */
885void splice_from_pipe_begin(struct splice_desc *sd)
886{
887 sd->num_spliced = 0;
888 sd->need_wakeup = false;
889}
890EXPORT_SYMBOL(splice_from_pipe_begin);
891
892/**
893 * splice_from_pipe_end - finish splicing from pipe
894 * @pipe: pipe to splice from
895 * @sd: information about the splice operation
896 *
897 * Description:
898 * This function will wake up pipe writers if necessary. It should
899 * be called after a loop containing splice_from_pipe_next() and
900 * splice_from_pipe_feed().
901 */
902void splice_from_pipe_end(struct pipe_inode_info *pipe, struct splice_desc *sd)
903{
904 if (sd->need_wakeup)
905 wakeup_pipe_writers(pipe);
906}
907EXPORT_SYMBOL(splice_from_pipe_end);
908
Jens Axboe932cc6d2007-06-21 13:10:21 +0200909/**
910 * __splice_from_pipe - splice data from a pipe to given actor
911 * @pipe: pipe to splice from
912 * @sd: information to @actor
913 * @actor: handler that splices the data
914 *
915 * Description:
916 * This function does little more than loop over the pipe and call
917 * @actor to do the actual moving of a single struct pipe_buffer to
918 * the desired destination. See pipe_to_file, pipe_to_sendpage, or
919 * pipe_to_user.
920 *
Jens Axboe83f91352006-04-02 23:05:09 +0200921 */
Jens Axboec66ab6f2007-06-12 21:17:17 +0200922ssize_t __splice_from_pipe(struct pipe_inode_info *pipe, struct splice_desc *sd,
923 splice_actor *actor)
Jens Axboe5274f052006-03-30 15:15:30 +0200924{
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200925 int ret;
Jens Axboe5274f052006-03-30 15:15:30 +0200926
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200927 splice_from_pipe_begin(sd);
928 do {
929 ret = splice_from_pipe_next(pipe, sd);
930 if (ret > 0)
931 ret = splice_from_pipe_feed(pipe, sd, actor);
932 } while (ret > 0);
933 splice_from_pipe_end(pipe, sd);
Jens Axboe5274f052006-03-30 15:15:30 +0200934
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200935 return sd->num_spliced ? sd->num_spliced : ret;
Jens Axboe5274f052006-03-30 15:15:30 +0200936}
Mark Fasheh40bee44e2007-03-21 13:11:02 +0100937EXPORT_SYMBOL(__splice_from_pipe);
Jens Axboe5274f052006-03-30 15:15:30 +0200938
Jens Axboe932cc6d2007-06-21 13:10:21 +0200939/**
940 * splice_from_pipe - splice data from a pipe to a file
941 * @pipe: pipe to splice from
942 * @out: file to splice to
943 * @ppos: position in @out
944 * @len: how many bytes to splice
945 * @flags: splice modifier flags
946 * @actor: handler that splices the data
947 *
948 * Description:
Miklos Szeredi29339702009-04-14 19:48:37 +0200949 * See __splice_from_pipe. This function locks the pipe inode,
Jens Axboe932cc6d2007-06-21 13:10:21 +0200950 * otherwise it's identical to __splice_from_pipe().
951 *
952 */
Mark Fasheh6da61802006-10-17 18:43:07 +0200953ssize_t splice_from_pipe(struct pipe_inode_info *pipe, struct file *out,
954 loff_t *ppos, size_t len, unsigned int flags,
955 splice_actor *actor)
956{
957 ssize_t ret;
Jens Axboec66ab6f2007-06-12 21:17:17 +0200958 struct splice_desc sd = {
959 .total_len = len,
960 .flags = flags,
961 .pos = *ppos,
Jens Axboe6a14b902007-06-14 13:08:55 +0200962 .u.file = out,
Jens Axboec66ab6f2007-06-12 21:17:17 +0200963 };
Mark Fasheh6da61802006-10-17 18:43:07 +0200964
Miklos Szeredi61e0d472009-04-14 19:48:41 +0200965 pipe_lock(pipe);
Jens Axboec66ab6f2007-06-12 21:17:17 +0200966 ret = __splice_from_pipe(pipe, &sd, actor);
Miklos Szeredi61e0d472009-04-14 19:48:41 +0200967 pipe_unlock(pipe);
Mark Fasheh6da61802006-10-17 18:43:07 +0200968
969 return ret;
970}
971
972/**
Jens Axboe83f91352006-04-02 23:05:09 +0200973 * generic_file_splice_write - splice data from a pipe to a file
Ingo Molnar3a326a22006-04-10 15:18:35 +0200974 * @pipe: pipe info
Jens Axboe83f91352006-04-02 23:05:09 +0200975 * @out: file to write to
Jens Axboe932cc6d2007-06-21 13:10:21 +0200976 * @ppos: position in @out
Jens Axboe83f91352006-04-02 23:05:09 +0200977 * @len: number of bytes to splice
978 * @flags: splice modifier flags
979 *
Jens Axboe932cc6d2007-06-21 13:10:21 +0200980 * Description:
981 * Will either move or copy pages (determined by @flags options) from
982 * the given pipe inode to the given file.
Jens Axboe83f91352006-04-02 23:05:09 +0200983 *
984 */
Ingo Molnar3a326a22006-04-10 15:18:35 +0200985ssize_t
986generic_file_splice_write(struct pipe_inode_info *pipe, struct file *out,
Jens Axboecbb7e572006-04-11 14:57:50 +0200987 loff_t *ppos, size_t len, unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +0200988{
Jens Axboe4f6f0bd2006-04-02 23:04:46 +0200989 struct address_space *mapping = out->f_mapping;
Jens Axboe8c34e2d2006-10-17 19:43:22 +0200990 struct inode *inode = mapping->host;
Miklos Szeredi7f3d4ee2008-05-07 09:22:39 +0200991 struct splice_desc sd = {
992 .total_len = len,
993 .flags = flags,
994 .pos = *ppos,
995 .u.file = out,
996 };
Ingo Molnar3a326a22006-04-10 15:18:35 +0200997 ssize_t ret;
Jens Axboe8c34e2d2006-10-17 19:43:22 +0200998
Jan Kara14da9202012-06-12 16:20:37 +0200999 sb_start_write(inode->i_sb);
1000
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001001 pipe_lock(pipe);
Miklos Szeredieb443e52009-04-14 19:48:38 +02001002
1003 splice_from_pipe_begin(&sd);
1004 do {
1005 ret = splice_from_pipe_next(pipe, &sd);
1006 if (ret <= 0)
1007 break;
1008
1009 mutex_lock_nested(&inode->i_mutex, I_MUTEX_CHILD);
1010 ret = file_remove_suid(out);
Miklos Szeredi723590e2009-08-15 08:43:22 +02001011 if (!ret) {
Josef Bacikc3b2da32012-03-26 09:59:21 -04001012 ret = file_update_time(out);
1013 if (!ret)
1014 ret = splice_from_pipe_feed(pipe, &sd,
1015 pipe_to_file);
Miklos Szeredi723590e2009-08-15 08:43:22 +02001016 }
Miklos Szeredieb443e52009-04-14 19:48:38 +02001017 mutex_unlock(&inode->i_mutex);
1018 } while (ret > 0);
1019 splice_from_pipe_end(pipe, &sd);
1020
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001021 pipe_unlock(pipe);
Miklos Szeredieb443e52009-04-14 19:48:38 +02001022
1023 if (sd.num_spliced)
1024 ret = sd.num_spliced;
1025
Jens Axboea4514eb2006-04-19 15:57:05 +02001026 if (ret > 0) {
Jens Axboe17ee4f42007-06-15 13:10:37 +02001027 unsigned long nr_pages;
Jan Kara148f9482009-08-17 19:52:36 +02001028 int err;
Jens Axboe17ee4f42007-06-15 13:10:37 +02001029
Jens Axboe17ee4f42007-06-15 13:10:37 +02001030 nr_pages = (ret + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
Jens Axboe4f6f0bd2006-04-02 23:04:46 +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;
Jens Axboe17ee4f42007-06-15 13:10:37 +02001037 balance_dirty_pages_ratelimited_nr(mapping, nr_pages);
Jens Axboe4f6f0bd2006-04-02 23:04:46 +02001038 }
Jan Kara14da9202012-06-12 16:20:37 +02001039 sb_end_write(inode->i_sb);
Jens Axboe4f6f0bd2006-04-02 23:04:46 +02001040
1041 return ret;
Jens Axboe5274f052006-03-30 15:15:30 +02001042}
1043
Jens Axboe059a8f32006-04-02 23:06:05 +02001044EXPORT_SYMBOL(generic_file_splice_write);
1045
Miklos Szeredib2858d72009-05-19 11:37:46 +02001046static int write_pipe_buf(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
1047 struct splice_desc *sd)
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +02001048{
Miklos Szeredib2858d72009-05-19 11:37:46 +02001049 int ret;
1050 void *data;
1051
Miklos Szeredib2858d72009-05-19 11:37:46 +02001052 data = buf->ops->map(pipe, buf, 0);
1053 ret = kernel_write(sd->u.file, data + buf->offset, sd->len, sd->pos);
1054 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
1118 return splice_write(pipe, out, ppos, len, flags);
Jens Axboe5274f052006-03-30 15:15:30 +02001119}
1120
Jens Axboe83f91352006-04-02 23:05:09 +02001121/*
1122 * Attempt to initiate a splice from a file to a pipe.
1123 */
Jens Axboecbb7e572006-04-11 14:57:50 +02001124static long do_splice_to(struct file *in, loff_t *ppos,
1125 struct pipe_inode_info *pipe, size_t len,
1126 unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +02001127{
Miklos Szeredi68181732009-05-07 15:37:36 +02001128 ssize_t (*splice_read)(struct file *, loff_t *,
1129 struct pipe_inode_info *, size_t, unsigned int);
Jens Axboe5274f052006-03-30 15:15:30 +02001130 int ret;
1131
Jens Axboe49570e92006-04-11 13:56:09 +02001132 if (unlikely(!(in->f_mode & FMODE_READ)))
Jens Axboe5274f052006-03-30 15:15:30 +02001133 return -EBADF;
1134
Jens Axboecbb7e572006-04-11 14:57:50 +02001135 ret = rw_verify_area(READ, in, ppos, len);
Jens Axboe5274f052006-03-30 15:15:30 +02001136 if (unlikely(ret < 0))
1137 return ret;
1138
Changli Gaocc56f7d2009-11-04 09:09:52 +01001139 if (in->f_op && in->f_op->splice_read)
1140 splice_read = in->f_op->splice_read;
1141 else
Miklos Szeredi68181732009-05-07 15:37:36 +02001142 splice_read = default_file_splice_read;
1143
1144 return splice_read(in, ppos, pipe, len, flags);
Jens Axboe5274f052006-03-30 15:15:30 +02001145}
1146
Jens Axboe932cc6d2007-06-21 13:10:21 +02001147/**
1148 * splice_direct_to_actor - splices data directly between two non-pipes
1149 * @in: file to splice from
1150 * @sd: actor information on where to splice to
1151 * @actor: handles the data splicing
1152 *
1153 * Description:
1154 * This is a special case helper to splice directly between two
1155 * points, without requiring an explicit pipe. Internally an allocated
Randy Dunlap79685b82007-07-27 08:08:51 +02001156 * pipe is cached in the process, and reused during the lifetime of
Jens Axboe932cc6d2007-06-21 13:10:21 +02001157 * that process.
1158 *
Jens Axboec66ab6f2007-06-12 21:17:17 +02001159 */
1160ssize_t splice_direct_to_actor(struct file *in, struct splice_desc *sd,
1161 splice_direct_actor *actor)
Jens Axboeb92ce552006-04-11 13:52:07 +02001162{
1163 struct pipe_inode_info *pipe;
1164 long ret, bytes;
1165 umode_t i_mode;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001166 size_t len;
1167 int i, flags;
Jens Axboeb92ce552006-04-11 13:52:07 +02001168
1169 /*
1170 * We require the input being a regular file, as we don't want to
1171 * randomly drop data for eg socket -> socket splicing. Use the
1172 * piped splicing for that!
1173 */
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001174 i_mode = in->f_path.dentry->d_inode->i_mode;
Jens Axboeb92ce552006-04-11 13:52:07 +02001175 if (unlikely(!S_ISREG(i_mode) && !S_ISBLK(i_mode)))
1176 return -EINVAL;
1177
1178 /*
1179 * neither in nor out is a pipe, setup an internal pipe attached to
1180 * 'out' and transfer the wanted data from 'in' to 'out' through that
1181 */
1182 pipe = current->splice_pipe;
Jens Axboe49570e92006-04-11 13:56:09 +02001183 if (unlikely(!pipe)) {
Jens Axboeb92ce552006-04-11 13:52:07 +02001184 pipe = alloc_pipe_info(NULL);
1185 if (!pipe)
1186 return -ENOMEM;
1187
1188 /*
1189 * We don't have an immediate reader, but we'll read the stuff
Jens Axboe00522fb2006-04-26 14:39:29 +02001190 * out of the pipe right after the splice_to_pipe(). So set
Jens Axboeb92ce552006-04-11 13:52:07 +02001191 * PIPE_READERS appropriately.
1192 */
1193 pipe->readers = 1;
1194
1195 current->splice_pipe = pipe;
1196 }
1197
1198 /*
Ingo Molnar73d62d82006-04-11 13:57:21 +02001199 * Do the splice.
Jens Axboeb92ce552006-04-11 13:52:07 +02001200 */
1201 ret = 0;
1202 bytes = 0;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001203 len = sd->total_len;
1204 flags = sd->flags;
1205
1206 /*
1207 * Don't block on output, we have to drain the direct pipe.
1208 */
1209 sd->flags &= ~SPLICE_F_NONBLOCK;
Jens Axboeb92ce552006-04-11 13:52:07 +02001210
1211 while (len) {
Jens Axboe51a92c02007-07-13 14:11:43 +02001212 size_t read_len;
Tom Zanussia82c53a2008-05-09 13:28:36 +02001213 loff_t pos = sd->pos, prev_pos = pos;
Jens Axboeb92ce552006-04-11 13:52:07 +02001214
Jens Axboebcd4f3a2007-07-16 14:41:49 +02001215 ret = do_splice_to(in, &pos, pipe, len, flags);
Jens Axboe51a92c02007-07-13 14:11:43 +02001216 if (unlikely(ret <= 0))
Jens Axboeb92ce552006-04-11 13:52:07 +02001217 goto out_release;
1218
1219 read_len = ret;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001220 sd->total_len = read_len;
Jens Axboeb92ce552006-04-11 13:52:07 +02001221
1222 /*
1223 * NOTE: nonblocking mode only applies to the input. We
1224 * must not do the output in nonblocking mode as then we
1225 * could get stuck data in the internal pipe:
1226 */
Jens Axboec66ab6f2007-06-12 21:17:17 +02001227 ret = actor(pipe, sd);
Tom Zanussia82c53a2008-05-09 13:28:36 +02001228 if (unlikely(ret <= 0)) {
1229 sd->pos = prev_pos;
Jens Axboeb92ce552006-04-11 13:52:07 +02001230 goto out_release;
Tom Zanussia82c53a2008-05-09 13:28:36 +02001231 }
Jens Axboeb92ce552006-04-11 13:52:07 +02001232
1233 bytes += ret;
1234 len -= ret;
Jens Axboebcd4f3a2007-07-16 14:41:49 +02001235 sd->pos = pos;
Jens Axboeb92ce552006-04-11 13:52:07 +02001236
Tom Zanussia82c53a2008-05-09 13:28:36 +02001237 if (ret < read_len) {
1238 sd->pos = prev_pos + ret;
Jens Axboe51a92c02007-07-13 14:11:43 +02001239 goto out_release;
Tom Zanussia82c53a2008-05-09 13:28:36 +02001240 }
Jens Axboeb92ce552006-04-11 13:52:07 +02001241 }
1242
Jens Axboe9e971982008-01-29 21:05:57 +01001243done:
Jens Axboeb92ce552006-04-11 13:52:07 +02001244 pipe->nrbufs = pipe->curbuf = 0;
Jens Axboe80848702008-01-30 12:24:48 +01001245 file_accessed(in);
Jens Axboeb92ce552006-04-11 13:52:07 +02001246 return bytes;
1247
1248out_release:
1249 /*
1250 * If we did an incomplete transfer we must release
1251 * the pipe buffers in question:
1252 */
Jens Axboe35f3d142010-05-20 10:43:18 +02001253 for (i = 0; i < pipe->buffers; i++) {
Jens Axboeb92ce552006-04-11 13:52:07 +02001254 struct pipe_buffer *buf = pipe->bufs + i;
1255
1256 if (buf->ops) {
1257 buf->ops->release(pipe, buf);
1258 buf->ops = NULL;
1259 }
1260 }
Jens Axboeb92ce552006-04-11 13:52:07 +02001261
Jens Axboe9e971982008-01-29 21:05:57 +01001262 if (!bytes)
1263 bytes = ret;
Jens Axboeb92ce552006-04-11 13:52:07 +02001264
Jens Axboe9e971982008-01-29 21:05:57 +01001265 goto done;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001266}
1267EXPORT_SYMBOL(splice_direct_to_actor);
1268
1269static int direct_splice_actor(struct pipe_inode_info *pipe,
1270 struct splice_desc *sd)
1271{
Jens Axboe6a14b902007-06-14 13:08:55 +02001272 struct file *file = sd->u.file;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001273
Changli Gao2cb4b052010-06-29 13:09:18 +02001274 return do_splice_from(pipe, file, &file->f_pos, sd->total_len,
1275 sd->flags);
Jens Axboec66ab6f2007-06-12 21:17:17 +02001276}
1277
Jens Axboe932cc6d2007-06-21 13:10:21 +02001278/**
1279 * do_splice_direct - splices data directly between two files
1280 * @in: file to splice from
1281 * @ppos: input file offset
1282 * @out: file to splice to
1283 * @len: number of bytes to splice
1284 * @flags: splice modifier flags
1285 *
1286 * Description:
1287 * For use by do_sendfile(). splice can easily emulate sendfile, but
1288 * doing it in the application would incur an extra system call
1289 * (splice in + splice out, as compared to just sendfile()). So this helper
1290 * can splice directly through a process-private pipe.
1291 *
1292 */
Jens Axboec66ab6f2007-06-12 21:17:17 +02001293long do_splice_direct(struct file *in, loff_t *ppos, struct file *out,
1294 size_t len, unsigned int flags)
1295{
1296 struct splice_desc sd = {
1297 .len = len,
1298 .total_len = len,
1299 .flags = flags,
1300 .pos = *ppos,
Jens Axboe6a14b902007-06-14 13:08:55 +02001301 .u.file = out,
Jens Axboec66ab6f2007-06-12 21:17:17 +02001302 };
Jens Axboe51a92c02007-07-13 14:11:43 +02001303 long ret;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001304
1305 ret = splice_direct_to_actor(in, &sd, direct_splice_actor);
Jens Axboe51a92c02007-07-13 14:11:43 +02001306 if (ret > 0)
Tom Zanussia82c53a2008-05-09 13:28:36 +02001307 *ppos = sd.pos;
Jens Axboe51a92c02007-07-13 14:11:43 +02001308
Jens Axboec66ab6f2007-06-12 21:17:17 +02001309 return ret;
Jens Axboeb92ce552006-04-11 13:52:07 +02001310}
1311
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001312static int splice_pipe_to_pipe(struct pipe_inode_info *ipipe,
1313 struct pipe_inode_info *opipe,
1314 size_t len, unsigned int flags);
Jens Axboeddac0d32006-11-04 12:49:32 +01001315
1316/*
Jens Axboe83f91352006-04-02 23:05:09 +02001317 * Determine where to splice to/from.
1318 */
Ingo Molnar529565d2006-04-10 15:18:58 +02001319static long do_splice(struct file *in, loff_t __user *off_in,
1320 struct file *out, loff_t __user *off_out,
1321 size_t len, unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +02001322{
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001323 struct pipe_inode_info *ipipe;
1324 struct pipe_inode_info *opipe;
Jens Axboecbb7e572006-04-11 14:57:50 +02001325 loff_t offset, *off;
Jens Axboea4514eb2006-04-19 15:57:05 +02001326 long ret;
Jens Axboe5274f052006-03-30 15:15:30 +02001327
Linus Torvalds71993e62010-11-28 13:56:09 -08001328 ipipe = get_pipe_info(in);
1329 opipe = get_pipe_info(out);
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001330
1331 if (ipipe && opipe) {
1332 if (off_in || off_out)
1333 return -ESPIPE;
1334
1335 if (!(in->f_mode & FMODE_READ))
1336 return -EBADF;
1337
1338 if (!(out->f_mode & FMODE_WRITE))
1339 return -EBADF;
1340
1341 /* Splicing to self would be fun, but... */
1342 if (ipipe == opipe)
1343 return -EINVAL;
1344
1345 return splice_pipe_to_pipe(ipipe, opipe, len, flags);
1346 }
1347
1348 if (ipipe) {
Ingo Molnar529565d2006-04-10 15:18:58 +02001349 if (off_in)
1350 return -ESPIPE;
Jens Axboeb92ce552006-04-11 13:52:07 +02001351 if (off_out) {
Changli Gao19c9a492010-06-29 13:10:36 +02001352 if (!(out->f_mode & FMODE_PWRITE))
Jens Axboeb92ce552006-04-11 13:52:07 +02001353 return -EINVAL;
Jens Axboecbb7e572006-04-11 14:57:50 +02001354 if (copy_from_user(&offset, off_out, sizeof(loff_t)))
Jens Axboeb92ce552006-04-11 13:52:07 +02001355 return -EFAULT;
Jens Axboecbb7e572006-04-11 14:57:50 +02001356 off = &offset;
1357 } else
1358 off = &out->f_pos;
Ingo Molnar529565d2006-04-10 15:18:58 +02001359
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001360 ret = do_splice_from(ipipe, out, off, len, flags);
Jens Axboea4514eb2006-04-19 15:57:05 +02001361
1362 if (off_out && copy_to_user(off_out, off, sizeof(loff_t)))
1363 ret = -EFAULT;
1364
1365 return ret;
Ingo Molnar529565d2006-04-10 15:18:58 +02001366 }
Jens Axboe5274f052006-03-30 15:15:30 +02001367
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001368 if (opipe) {
Ingo Molnar529565d2006-04-10 15:18:58 +02001369 if (off_out)
1370 return -ESPIPE;
Jens Axboeb92ce552006-04-11 13:52:07 +02001371 if (off_in) {
Changli Gao19c9a492010-06-29 13:10:36 +02001372 if (!(in->f_mode & FMODE_PREAD))
Jens Axboeb92ce552006-04-11 13:52:07 +02001373 return -EINVAL;
Jens Axboecbb7e572006-04-11 14:57:50 +02001374 if (copy_from_user(&offset, off_in, sizeof(loff_t)))
Jens Axboeb92ce552006-04-11 13:52:07 +02001375 return -EFAULT;
Jens Axboecbb7e572006-04-11 14:57:50 +02001376 off = &offset;
1377 } else
1378 off = &in->f_pos;
Ingo Molnar529565d2006-04-10 15:18:58 +02001379
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001380 ret = do_splice_to(in, off, opipe, len, flags);
Jens Axboea4514eb2006-04-19 15:57:05 +02001381
1382 if (off_in && copy_to_user(off_in, off, sizeof(loff_t)))
1383 ret = -EFAULT;
1384
1385 return ret;
Ingo Molnar529565d2006-04-10 15:18:58 +02001386 }
Jens Axboe5274f052006-03-30 15:15:30 +02001387
1388 return -EINVAL;
1389}
1390
Jens Axboe912d35f2006-04-26 10:59:21 +02001391/*
1392 * Map an iov into an array of pages and offset/length tupples. With the
1393 * partial_page structure, we can map several non-contiguous ranges into
1394 * our ones pages[] map instead of splitting that operation into pieces.
1395 * Could easily be exported as a generic helper for other users, in which
1396 * case one would probably want to add a 'max_nr_pages' parameter as well.
1397 */
1398static int get_iovec_page_array(const struct iovec __user *iov,
1399 unsigned int nr_vecs, struct page **pages,
Eric Dumazetbd1a68b2012-04-04 11:08:51 +02001400 struct partial_page *partial, bool aligned,
Jens Axboe35f3d142010-05-20 10:43:18 +02001401 unsigned int pipe_buffers)
Jens Axboe912d35f2006-04-26 10:59:21 +02001402{
1403 int buffers = 0, error = 0;
1404
Jens Axboe912d35f2006-04-26 10:59:21 +02001405 while (nr_vecs) {
1406 unsigned long off, npages;
Linus Torvalds75723952007-10-01 13:17:28 -07001407 struct iovec entry;
Jens Axboe912d35f2006-04-26 10:59:21 +02001408 void __user *base;
1409 size_t len;
1410 int i;
1411
Linus Torvalds75723952007-10-01 13:17:28 -07001412 error = -EFAULT;
Nick Pigginbc40d732008-07-25 19:45:26 -07001413 if (copy_from_user(&entry, iov, sizeof(entry)))
Jens Axboe912d35f2006-04-26 10:59:21 +02001414 break;
Linus Torvalds75723952007-10-01 13:17:28 -07001415
1416 base = entry.iov_base;
1417 len = entry.iov_len;
Jens Axboe912d35f2006-04-26 10:59:21 +02001418
1419 /*
1420 * Sanity check this iovec. 0 read succeeds.
1421 */
Linus Torvalds75723952007-10-01 13:17:28 -07001422 error = 0;
Jens Axboe912d35f2006-04-26 10:59:21 +02001423 if (unlikely(!len))
1424 break;
1425 error = -EFAULT;
Bastian Blank712a30e2008-02-10 16:47:57 +02001426 if (!access_ok(VERIFY_READ, base, len))
Jens Axboe912d35f2006-04-26 10:59:21 +02001427 break;
1428
1429 /*
1430 * Get this base offset and number of pages, then map
1431 * in the user pages.
1432 */
1433 off = (unsigned long) base & ~PAGE_MASK;
Jens Axboe7afa6fd2006-05-01 20:02:33 +02001434
1435 /*
1436 * If asked for alignment, the offset must be zero and the
1437 * length a multiple of the PAGE_SIZE.
1438 */
1439 error = -EINVAL;
1440 if (aligned && (off || len & ~PAGE_MASK))
1441 break;
1442
Jens Axboe912d35f2006-04-26 10:59:21 +02001443 npages = (off + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
Jens Axboe35f3d142010-05-20 10:43:18 +02001444 if (npages > pipe_buffers - buffers)
1445 npages = pipe_buffers - buffers;
Jens Axboe912d35f2006-04-26 10:59:21 +02001446
Nick Pigginbc40d732008-07-25 19:45:26 -07001447 error = get_user_pages_fast((unsigned long)base, npages,
1448 0, &pages[buffers]);
Jens Axboe912d35f2006-04-26 10:59:21 +02001449
1450 if (unlikely(error <= 0))
1451 break;
1452
1453 /*
1454 * Fill this contiguous range into the partial page map.
1455 */
1456 for (i = 0; i < error; i++) {
Jens Axboe75914892006-05-02 12:57:18 +02001457 const int plen = min_t(size_t, len, PAGE_SIZE - off);
Jens Axboe912d35f2006-04-26 10:59:21 +02001458
1459 partial[buffers].offset = off;
1460 partial[buffers].len = plen;
1461
1462 off = 0;
1463 len -= plen;
1464 buffers++;
1465 }
1466
1467 /*
1468 * We didn't complete this iov, stop here since it probably
1469 * means we have to move some of this into a pipe to
1470 * be able to continue.
1471 */
1472 if (len)
1473 break;
1474
1475 /*
1476 * Don't continue if we mapped fewer pages than we asked for,
1477 * or if we mapped the max number of pages that we have
1478 * room for.
1479 */
Jens Axboe35f3d142010-05-20 10:43:18 +02001480 if (error < npages || buffers == pipe_buffers)
Jens Axboe912d35f2006-04-26 10:59:21 +02001481 break;
1482
1483 nr_vecs--;
1484 iov++;
1485 }
1486
Jens Axboe912d35f2006-04-26 10:59:21 +02001487 if (buffers)
1488 return buffers;
1489
1490 return error;
1491}
1492
Jens Axboe6a14b902007-06-14 13:08:55 +02001493static int pipe_to_user(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
1494 struct splice_desc *sd)
1495{
1496 char *src;
1497 int ret;
1498
Jens Axboe6a14b902007-06-14 13:08:55 +02001499 /*
1500 * See if we can use the atomic maps, by prefaulting in the
1501 * pages and doing an atomic copy
1502 */
1503 if (!fault_in_pages_writeable(sd->u.userptr, sd->len)) {
1504 src = buf->ops->map(pipe, buf, 1);
1505 ret = __copy_to_user_inatomic(sd->u.userptr, src + buf->offset,
1506 sd->len);
1507 buf->ops->unmap(pipe, buf, src);
1508 if (!ret) {
1509 ret = sd->len;
1510 goto out;
1511 }
1512 }
1513
1514 /*
1515 * No dice, use slow non-atomic map and copy
1516 */
1517 src = buf->ops->map(pipe, buf, 0);
1518
1519 ret = sd->len;
1520 if (copy_to_user(sd->u.userptr, src + buf->offset, sd->len))
1521 ret = -EFAULT;
1522
Jens Axboe6866bef2007-10-16 10:01:29 +02001523 buf->ops->unmap(pipe, buf, src);
Jens Axboe6a14b902007-06-14 13:08:55 +02001524out:
1525 if (ret > 0)
1526 sd->u.userptr += ret;
Jens Axboe6a14b902007-06-14 13:08:55 +02001527 return ret;
1528}
1529
1530/*
1531 * For lack of a better implementation, implement vmsplice() to userspace
1532 * as a simple copy of the pipes pages to the user iov.
1533 */
1534static long vmsplice_to_user(struct file *file, const struct iovec __user *iov,
1535 unsigned long nr_segs, unsigned int flags)
1536{
1537 struct pipe_inode_info *pipe;
1538 struct splice_desc sd;
1539 ssize_t size;
1540 int error;
1541 long ret;
1542
Linus Torvalds71993e62010-11-28 13:56:09 -08001543 pipe = get_pipe_info(file);
Jens Axboe6a14b902007-06-14 13:08:55 +02001544 if (!pipe)
1545 return -EBADF;
1546
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001547 pipe_lock(pipe);
Jens Axboe6a14b902007-06-14 13:08:55 +02001548
1549 error = ret = 0;
1550 while (nr_segs) {
1551 void __user *base;
1552 size_t len;
1553
1554 /*
1555 * Get user address base and length for this iovec.
1556 */
1557 error = get_user(base, &iov->iov_base);
1558 if (unlikely(error))
1559 break;
1560 error = get_user(len, &iov->iov_len);
1561 if (unlikely(error))
1562 break;
1563
1564 /*
1565 * Sanity check this iovec. 0 read succeeds.
1566 */
1567 if (unlikely(!len))
1568 break;
1569 if (unlikely(!base)) {
1570 error = -EFAULT;
1571 break;
1572 }
1573
Jens Axboe88119302008-02-08 08:49:14 -08001574 if (unlikely(!access_ok(VERIFY_WRITE, base, len))) {
1575 error = -EFAULT;
1576 break;
1577 }
1578
Jens Axboe6a14b902007-06-14 13:08:55 +02001579 sd.len = 0;
1580 sd.total_len = len;
1581 sd.flags = flags;
1582 sd.u.userptr = base;
1583 sd.pos = 0;
1584
1585 size = __splice_from_pipe(pipe, &sd, pipe_to_user);
1586 if (size < 0) {
1587 if (!ret)
1588 ret = size;
1589
1590 break;
1591 }
1592
1593 ret += size;
1594
1595 if (size < len)
1596 break;
1597
1598 nr_segs--;
1599 iov++;
1600 }
1601
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001602 pipe_unlock(pipe);
Jens Axboe6a14b902007-06-14 13:08:55 +02001603
1604 if (!ret)
1605 ret = error;
1606
1607 return ret;
1608}
1609
Jens Axboe912d35f2006-04-26 10:59:21 +02001610/*
1611 * vmsplice splices a user address range into a pipe. It can be thought of
1612 * as splice-from-memory, where the regular splice is splice-from-file (or
1613 * to file). In both cases the output is a pipe, naturally.
Jens Axboe912d35f2006-04-26 10:59:21 +02001614 */
Jens Axboe6a14b902007-06-14 13:08:55 +02001615static long vmsplice_to_pipe(struct file *file, const struct iovec __user *iov,
1616 unsigned long nr_segs, unsigned int flags)
Jens Axboe912d35f2006-04-26 10:59:21 +02001617{
Jens Axboeddac0d32006-11-04 12:49:32 +01001618 struct pipe_inode_info *pipe;
Jens Axboe35f3d142010-05-20 10:43:18 +02001619 struct page *pages[PIPE_DEF_BUFFERS];
1620 struct partial_page partial[PIPE_DEF_BUFFERS];
Jens Axboe912d35f2006-04-26 10:59:21 +02001621 struct splice_pipe_desc spd = {
1622 .pages = pages,
1623 .partial = partial,
Eric Dumazet047fe362012-06-12 15:24:40 +02001624 .nr_pages_max = PIPE_DEF_BUFFERS,
Jens Axboe912d35f2006-04-26 10:59:21 +02001625 .flags = flags,
1626 .ops = &user_page_pipe_buf_ops,
Jens Axboebbdfc2f2007-11-06 23:29:47 -08001627 .spd_release = spd_release_page,
Jens Axboe912d35f2006-04-26 10:59:21 +02001628 };
Jens Axboe35f3d142010-05-20 10:43:18 +02001629 long ret;
Jens Axboe912d35f2006-04-26 10:59:21 +02001630
Linus Torvalds71993e62010-11-28 13:56:09 -08001631 pipe = get_pipe_info(file);
Jens Axboeddac0d32006-11-04 12:49:32 +01001632 if (!pipe)
Jens Axboe912d35f2006-04-26 10:59:21 +02001633 return -EBADF;
Jens Axboe912d35f2006-04-26 10:59:21 +02001634
Jens Axboe35f3d142010-05-20 10:43:18 +02001635 if (splice_grow_spd(pipe, &spd))
1636 return -ENOMEM;
Jens Axboe912d35f2006-04-26 10:59:21 +02001637
Jens Axboe35f3d142010-05-20 10:43:18 +02001638 spd.nr_pages = get_iovec_page_array(iov, nr_segs, spd.pages,
Eric Dumazetbd1a68b2012-04-04 11:08:51 +02001639 spd.partial, false,
Eric Dumazet047fe362012-06-12 15:24:40 +02001640 spd.nr_pages_max);
Jens Axboe35f3d142010-05-20 10:43:18 +02001641 if (spd.nr_pages <= 0)
1642 ret = spd.nr_pages;
1643 else
1644 ret = splice_to_pipe(pipe, &spd);
1645
Eric Dumazet047fe362012-06-12 15:24:40 +02001646 splice_shrink_spd(&spd);
Jens Axboe35f3d142010-05-20 10:43:18 +02001647 return ret;
Jens Axboe912d35f2006-04-26 10:59:21 +02001648}
1649
Jens Axboe6a14b902007-06-14 13:08:55 +02001650/*
1651 * Note that vmsplice only really supports true splicing _from_ user memory
1652 * to a pipe, not the other way around. Splicing from user memory is a simple
1653 * operation that can be supported without any funky alignment restrictions
1654 * or nasty vm tricks. We simply map in the user memory and fill them into
1655 * a pipe. The reverse isn't quite as easy, though. There are two possible
1656 * solutions for that:
1657 *
1658 * - memcpy() the data internally, at which point we might as well just
1659 * do a regular read() on the buffer anyway.
1660 * - Lots of nasty vm tricks, that are neither fast nor flexible (it
1661 * has restriction limitations on both ends of the pipe).
1662 *
1663 * Currently we punt and implement it as a normal copy, see pipe_to_user().
1664 *
1665 */
Heiko Carstens836f92a2009-01-14 14:14:33 +01001666SYSCALL_DEFINE4(vmsplice, int, fd, const struct iovec __user *, iov,
1667 unsigned long, nr_segs, unsigned int, flags)
Jens Axboe912d35f2006-04-26 10:59:21 +02001668{
Al Viro2903ff02012-08-28 12:52:22 -04001669 struct fd f;
Jens Axboe912d35f2006-04-26 10:59:21 +02001670 long error;
Jens Axboe912d35f2006-04-26 10:59:21 +02001671
Jens Axboe6a14b902007-06-14 13:08:55 +02001672 if (unlikely(nr_segs > UIO_MAXIOV))
1673 return -EINVAL;
1674 else if (unlikely(!nr_segs))
1675 return 0;
1676
Jens Axboe912d35f2006-04-26 10:59:21 +02001677 error = -EBADF;
Al Viro2903ff02012-08-28 12:52:22 -04001678 f = fdget(fd);
1679 if (f.file) {
1680 if (f.file->f_mode & FMODE_WRITE)
1681 error = vmsplice_to_pipe(f.file, iov, nr_segs, flags);
1682 else if (f.file->f_mode & FMODE_READ)
1683 error = vmsplice_to_user(f.file, iov, nr_segs, flags);
Jens Axboe912d35f2006-04-26 10:59:21 +02001684
Al Viro2903ff02012-08-28 12:52:22 -04001685 fdput(f);
Jens Axboe912d35f2006-04-26 10:59:21 +02001686 }
1687
1688 return error;
1689}
1690
Heiko Carstens836f92a2009-01-14 14:14:33 +01001691SYSCALL_DEFINE6(splice, int, fd_in, loff_t __user *, off_in,
1692 int, fd_out, loff_t __user *, off_out,
1693 size_t, len, unsigned int, flags)
Jens Axboe5274f052006-03-30 15:15:30 +02001694{
Al Viro2903ff02012-08-28 12:52:22 -04001695 struct fd in, out;
Jens Axboe5274f052006-03-30 15:15:30 +02001696 long error;
Jens Axboe5274f052006-03-30 15:15:30 +02001697
1698 if (unlikely(!len))
1699 return 0;
1700
1701 error = -EBADF;
Al Viro2903ff02012-08-28 12:52:22 -04001702 in = fdget(fd_in);
1703 if (in.file) {
1704 if (in.file->f_mode & FMODE_READ) {
1705 out = fdget(fd_out);
1706 if (out.file) {
1707 if (out.file->f_mode & FMODE_WRITE)
1708 error = do_splice(in.file, off_in,
1709 out.file, off_out,
Ingo Molnar529565d2006-04-10 15:18:58 +02001710 len, flags);
Al Viro2903ff02012-08-28 12:52:22 -04001711 fdput(out);
Jens Axboe5274f052006-03-30 15:15:30 +02001712 }
1713 }
Al Viro2903ff02012-08-28 12:52:22 -04001714 fdput(in);
Jens Axboe5274f052006-03-30 15:15:30 +02001715 }
Jens Axboe5274f052006-03-30 15:15:30 +02001716 return error;
1717}
Jens Axboe70524492006-04-11 15:51:17 +02001718
1719/*
Jens Axboeaadd06e2006-07-10 11:00:01 +02001720 * Make sure there's data to read. Wait for input if we can, otherwise
1721 * return an appropriate error.
1722 */
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001723static int ipipe_prep(struct pipe_inode_info *pipe, unsigned int flags)
Jens Axboeaadd06e2006-07-10 11:00:01 +02001724{
1725 int ret;
1726
1727 /*
1728 * Check ->nrbufs without the inode lock first. This function
1729 * is speculative anyways, so missing one is ok.
1730 */
1731 if (pipe->nrbufs)
1732 return 0;
1733
1734 ret = 0;
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001735 pipe_lock(pipe);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001736
1737 while (!pipe->nrbufs) {
1738 if (signal_pending(current)) {
1739 ret = -ERESTARTSYS;
1740 break;
1741 }
1742 if (!pipe->writers)
1743 break;
1744 if (!pipe->waiting_writers) {
1745 if (flags & SPLICE_F_NONBLOCK) {
1746 ret = -EAGAIN;
1747 break;
1748 }
1749 }
1750 pipe_wait(pipe);
1751 }
1752
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001753 pipe_unlock(pipe);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001754 return ret;
1755}
1756
1757/*
1758 * Make sure there's writeable room. Wait for room if we can, otherwise
1759 * return an appropriate error.
1760 */
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001761static int opipe_prep(struct pipe_inode_info *pipe, unsigned int flags)
Jens Axboeaadd06e2006-07-10 11:00:01 +02001762{
1763 int ret;
1764
1765 /*
1766 * Check ->nrbufs without the inode lock first. This function
1767 * is speculative anyways, so missing one is ok.
1768 */
Jens Axboe35f3d142010-05-20 10:43:18 +02001769 if (pipe->nrbufs < pipe->buffers)
Jens Axboeaadd06e2006-07-10 11:00:01 +02001770 return 0;
1771
1772 ret = 0;
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001773 pipe_lock(pipe);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001774
Jens Axboe35f3d142010-05-20 10:43:18 +02001775 while (pipe->nrbufs >= pipe->buffers) {
Jens Axboeaadd06e2006-07-10 11:00:01 +02001776 if (!pipe->readers) {
1777 send_sig(SIGPIPE, current, 0);
1778 ret = -EPIPE;
1779 break;
1780 }
1781 if (flags & SPLICE_F_NONBLOCK) {
1782 ret = -EAGAIN;
1783 break;
1784 }
1785 if (signal_pending(current)) {
1786 ret = -ERESTARTSYS;
1787 break;
1788 }
1789 pipe->waiting_writers++;
1790 pipe_wait(pipe);
1791 pipe->waiting_writers--;
1792 }
1793
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001794 pipe_unlock(pipe);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001795 return ret;
1796}
1797
1798/*
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001799 * Splice contents of ipipe to opipe.
1800 */
1801static int splice_pipe_to_pipe(struct pipe_inode_info *ipipe,
1802 struct pipe_inode_info *opipe,
1803 size_t len, unsigned int flags)
1804{
1805 struct pipe_buffer *ibuf, *obuf;
1806 int ret = 0, nbuf;
1807 bool input_wakeup = false;
1808
1809
1810retry:
1811 ret = ipipe_prep(ipipe, flags);
1812 if (ret)
1813 return ret;
1814
1815 ret = opipe_prep(opipe, flags);
1816 if (ret)
1817 return ret;
1818
1819 /*
1820 * Potential ABBA deadlock, work around it by ordering lock
1821 * grabbing by pipe info address. Otherwise two different processes
1822 * could deadlock (one doing tee from A -> B, the other from B -> A).
1823 */
1824 pipe_double_lock(ipipe, opipe);
1825
1826 do {
1827 if (!opipe->readers) {
1828 send_sig(SIGPIPE, current, 0);
1829 if (!ret)
1830 ret = -EPIPE;
1831 break;
1832 }
1833
1834 if (!ipipe->nrbufs && !ipipe->writers)
1835 break;
1836
1837 /*
1838 * Cannot make any progress, because either the input
1839 * pipe is empty or the output pipe is full.
1840 */
Jens Axboe35f3d142010-05-20 10:43:18 +02001841 if (!ipipe->nrbufs || opipe->nrbufs >= opipe->buffers) {
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001842 /* Already processed some buffers, break */
1843 if (ret)
1844 break;
1845
1846 if (flags & SPLICE_F_NONBLOCK) {
1847 ret = -EAGAIN;
1848 break;
1849 }
1850
1851 /*
1852 * We raced with another reader/writer and haven't
1853 * managed to process any buffers. A zero return
1854 * value means EOF, so retry instead.
1855 */
1856 pipe_unlock(ipipe);
1857 pipe_unlock(opipe);
1858 goto retry;
1859 }
1860
1861 ibuf = ipipe->bufs + ipipe->curbuf;
Jens Axboe35f3d142010-05-20 10:43:18 +02001862 nbuf = (opipe->curbuf + opipe->nrbufs) & (opipe->buffers - 1);
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001863 obuf = opipe->bufs + nbuf;
1864
1865 if (len >= ibuf->len) {
1866 /*
1867 * Simply move the whole buffer from ipipe to opipe
1868 */
1869 *obuf = *ibuf;
1870 ibuf->ops = NULL;
1871 opipe->nrbufs++;
Jens Axboe35f3d142010-05-20 10:43:18 +02001872 ipipe->curbuf = (ipipe->curbuf + 1) & (ipipe->buffers - 1);
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001873 ipipe->nrbufs--;
1874 input_wakeup = true;
1875 } else {
1876 /*
1877 * Get a reference to this pipe buffer,
1878 * so we can copy the contents over.
1879 */
1880 ibuf->ops->get(ipipe, ibuf);
1881 *obuf = *ibuf;
1882
1883 /*
1884 * Don't inherit the gift flag, we need to
1885 * prevent multiple steals of this page.
1886 */
1887 obuf->flags &= ~PIPE_BUF_FLAG_GIFT;
1888
1889 obuf->len = len;
1890 opipe->nrbufs++;
1891 ibuf->offset += obuf->len;
1892 ibuf->len -= obuf->len;
1893 }
1894 ret += obuf->len;
1895 len -= obuf->len;
1896 } while (len);
1897
1898 pipe_unlock(ipipe);
1899 pipe_unlock(opipe);
1900
1901 /*
1902 * If we put data in the output pipe, wakeup any potential readers.
1903 */
Namhyung Kim825cdcb2011-05-23 19:58:53 +02001904 if (ret > 0)
1905 wakeup_pipe_readers(opipe);
1906
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001907 if (input_wakeup)
1908 wakeup_pipe_writers(ipipe);
1909
1910 return ret;
1911}
1912
1913/*
Jens Axboe70524492006-04-11 15:51:17 +02001914 * Link contents of ipipe to opipe.
1915 */
1916static int link_pipe(struct pipe_inode_info *ipipe,
1917 struct pipe_inode_info *opipe,
1918 size_t len, unsigned int flags)
1919{
1920 struct pipe_buffer *ibuf, *obuf;
Jens Axboeaadd06e2006-07-10 11:00:01 +02001921 int ret = 0, i = 0, nbuf;
Jens Axboe70524492006-04-11 15:51:17 +02001922
1923 /*
1924 * Potential ABBA deadlock, work around it by ordering lock
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001925 * grabbing by pipe info address. Otherwise two different processes
Jens Axboe70524492006-04-11 15:51:17 +02001926 * could deadlock (one doing tee from A -> B, the other from B -> A).
1927 */
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001928 pipe_double_lock(ipipe, opipe);
Jens Axboe70524492006-04-11 15:51:17 +02001929
Jens Axboeaadd06e2006-07-10 11:00:01 +02001930 do {
Jens Axboe70524492006-04-11 15:51:17 +02001931 if (!opipe->readers) {
1932 send_sig(SIGPIPE, current, 0);
1933 if (!ret)
1934 ret = -EPIPE;
1935 break;
1936 }
Jens Axboe70524492006-04-11 15:51:17 +02001937
1938 /*
Jens Axboeaadd06e2006-07-10 11:00:01 +02001939 * If we have iterated all input buffers or ran out of
1940 * output room, break.
Jens Axboe70524492006-04-11 15:51:17 +02001941 */
Jens Axboe35f3d142010-05-20 10:43:18 +02001942 if (i >= ipipe->nrbufs || opipe->nrbufs >= opipe->buffers)
Jens Axboe70524492006-04-11 15:51:17 +02001943 break;
Jens Axboeaadd06e2006-07-10 11:00:01 +02001944
Jens Axboe35f3d142010-05-20 10:43:18 +02001945 ibuf = ipipe->bufs + ((ipipe->curbuf + i) & (ipipe->buffers-1));
1946 nbuf = (opipe->curbuf + opipe->nrbufs) & (opipe->buffers - 1);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001947
Jens Axboe2a27250e2006-04-19 15:56:40 +02001948 /*
Jens Axboeaadd06e2006-07-10 11:00:01 +02001949 * Get a reference to this pipe buffer,
1950 * so we can copy the contents over.
Jens Axboe2a27250e2006-04-19 15:56:40 +02001951 */
Jens Axboeaadd06e2006-07-10 11:00:01 +02001952 ibuf->ops->get(ipipe, ibuf);
Jens Axboe70524492006-04-11 15:51:17 +02001953
Jens Axboeaadd06e2006-07-10 11:00:01 +02001954 obuf = opipe->bufs + nbuf;
1955 *obuf = *ibuf;
Jens Axboe70524492006-04-11 15:51:17 +02001956
Jens Axboeaadd06e2006-07-10 11:00:01 +02001957 /*
1958 * Don't inherit the gift flag, we need to
1959 * prevent multiple steals of this page.
1960 */
1961 obuf->flags &= ~PIPE_BUF_FLAG_GIFT;
1962
1963 if (obuf->len > len)
1964 obuf->len = len;
1965
1966 opipe->nrbufs++;
1967 ret += obuf->len;
1968 len -= obuf->len;
1969 i++;
1970 } while (len);
Jens Axboe70524492006-04-11 15:51:17 +02001971
Jens Axboe02cf01a2008-02-20 10:34:51 +01001972 /*
1973 * return EAGAIN if we have the potential of some data in the
1974 * future, otherwise just return 0
1975 */
1976 if (!ret && ipipe->waiting_writers && (flags & SPLICE_F_NONBLOCK))
1977 ret = -EAGAIN;
1978
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001979 pipe_unlock(ipipe);
1980 pipe_unlock(opipe);
Jens Axboe70524492006-04-11 15:51:17 +02001981
Jens Axboeaadd06e2006-07-10 11:00:01 +02001982 /*
1983 * If we put data in the output pipe, wakeup any potential readers.
1984 */
Namhyung Kim825cdcb2011-05-23 19:58:53 +02001985 if (ret > 0)
1986 wakeup_pipe_readers(opipe);
Jens Axboe70524492006-04-11 15:51:17 +02001987
1988 return ret;
1989}
1990
1991/*
1992 * This is a tee(1) implementation that works on pipes. It doesn't copy
1993 * any data, it simply references the 'in' pages on the 'out' pipe.
1994 * The 'flags' used are the SPLICE_F_* variants, currently the only
1995 * applicable one is SPLICE_F_NONBLOCK.
1996 */
1997static long do_tee(struct file *in, struct file *out, size_t len,
1998 unsigned int flags)
1999{
Linus Torvalds71993e62010-11-28 13:56:09 -08002000 struct pipe_inode_info *ipipe = get_pipe_info(in);
2001 struct pipe_inode_info *opipe = get_pipe_info(out);
Jens Axboeaadd06e2006-07-10 11:00:01 +02002002 int ret = -EINVAL;
Jens Axboe70524492006-04-11 15:51:17 +02002003
2004 /*
Jens Axboeaadd06e2006-07-10 11:00:01 +02002005 * Duplicate the contents of ipipe to opipe without actually
2006 * copying the data.
Jens Axboe70524492006-04-11 15:51:17 +02002007 */
Jens Axboeaadd06e2006-07-10 11:00:01 +02002008 if (ipipe && opipe && ipipe != opipe) {
2009 /*
2010 * Keep going, unless we encounter an error. The ipipe/opipe
2011 * ordering doesn't really matter.
2012 */
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02002013 ret = ipipe_prep(ipipe, flags);
Jens Axboeaadd06e2006-07-10 11:00:01 +02002014 if (!ret) {
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02002015 ret = opipe_prep(opipe, flags);
Jens Axboe02cf01a2008-02-20 10:34:51 +01002016 if (!ret)
Jens Axboeaadd06e2006-07-10 11:00:01 +02002017 ret = link_pipe(ipipe, opipe, len, flags);
Jens Axboeaadd06e2006-07-10 11:00:01 +02002018 }
2019 }
Jens Axboe70524492006-04-11 15:51:17 +02002020
Jens Axboeaadd06e2006-07-10 11:00:01 +02002021 return ret;
Jens Axboe70524492006-04-11 15:51:17 +02002022}
2023
Heiko Carstens836f92a2009-01-14 14:14:33 +01002024SYSCALL_DEFINE4(tee, int, fdin, int, fdout, size_t, len, unsigned int, flags)
Jens Axboe70524492006-04-11 15:51:17 +02002025{
Al Viro2903ff02012-08-28 12:52:22 -04002026 struct fd in;
2027 int error;
Jens Axboe70524492006-04-11 15:51:17 +02002028
2029 if (unlikely(!len))
2030 return 0;
2031
2032 error = -EBADF;
Al Viro2903ff02012-08-28 12:52:22 -04002033 in = fdget(fdin);
2034 if (in.file) {
2035 if (in.file->f_mode & FMODE_READ) {
2036 struct fd out = fdget(fdout);
2037 if (out.file) {
2038 if (out.file->f_mode & FMODE_WRITE)
2039 error = do_tee(in.file, out.file,
2040 len, flags);
2041 fdput(out);
Jens Axboe70524492006-04-11 15:51:17 +02002042 }
2043 }
Al Viro2903ff02012-08-28 12:52:22 -04002044 fdput(in);
Jens Axboe70524492006-04-11 15:51:17 +02002045 }
2046
2047 return error;
2048}