blob: aa866d309695497c1c8925cb30c8af6884f20aaf [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>
28#include <linux/buffer_head.h>
Jeff Garzika0f06782006-03-30 23:06:13 -050029#include <linux/module.h>
Jens Axboe4f6f0bd2006-04-02 23:04:46 +020030#include <linux/syscalls.h>
Jens Axboe912d35f2006-04-26 10:59:21 +020031#include <linux/uio.h>
James Morris29ce2052007-07-13 11:44:32 +020032#include <linux/security.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090033#include <linux/gfp.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 Axboecac36bb2007-06-14 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
Eric Dumazetd4c3cca2006-12-13 00:34:04 -0800135static const 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 Axboecac36bb2007-06-14 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 Axboecac36bb2007-06-14 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
Jens Axboebbdfc2f2007-11-06 23:29:47 -0800267static void spd_release_page(struct splice_pipe_desc *spd, unsigned int i)
268{
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 */
276int splice_grow_spd(struct pipe_inode_info *pipe, struct splice_pipe_desc *spd)
277{
278 if (pipe->buffers <= PIPE_DEF_BUFFERS)
279 return 0;
280
281 spd->pages = kmalloc(pipe->buffers * sizeof(struct page *), GFP_KERNEL);
282 spd->partial = kmalloc(pipe->buffers * sizeof(struct partial_page), GFP_KERNEL);
283
284 if (spd->pages && spd->partial)
285 return 0;
286
287 kfree(spd->pages);
288 kfree(spd->partial);
289 return -ENOMEM;
290}
291
292void splice_shrink_spd(struct pipe_inode_info *pipe,
293 struct splice_pipe_desc *spd)
294{
295 if (pipe->buffers <= PIPE_DEF_BUFFERS)
296 return;
297
298 kfree(spd->pages);
299 kfree(spd->partial);
300}
301
Ingo Molnar3a326a22006-04-10 15:18:35 +0200302static int
Jens Axboecbb7e572006-04-11 14:57:50 +0200303__generic_file_splice_read(struct file *in, loff_t *ppos,
304 struct pipe_inode_info *pipe, size_t len,
305 unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +0200306{
307 struct address_space *mapping = in->f_mapping;
Fengguang Wud8983912007-07-19 01:48:06 -0700308 unsigned int loff, nr_pages, req_pages;
Jens Axboe35f3d142010-05-20 10:43:18 +0200309 struct page *pages[PIPE_DEF_BUFFERS];
310 struct partial_page partial[PIPE_DEF_BUFFERS];
Jens Axboe5274f052006-03-30 15:15:30 +0200311 struct page *page;
Jens Axboe91ad66e2006-04-19 15:55:10 +0200312 pgoff_t index, end_index;
313 loff_t isize;
Jens Axboeeb207962006-04-27 11:05:22 +0200314 int error, page_nr;
Jens Axboe912d35f2006-04-26 10:59:21 +0200315 struct splice_pipe_desc spd = {
316 .pages = pages,
317 .partial = partial,
318 .flags = flags,
319 .ops = &page_cache_pipe_buf_ops,
Jens Axboebbdfc2f2007-11-06 23:29:47 -0800320 .spd_release = spd_release_page,
Jens Axboe912d35f2006-04-26 10:59:21 +0200321 };
Jens Axboe5274f052006-03-30 15:15:30 +0200322
Jens Axboe35f3d142010-05-20 10:43:18 +0200323 if (splice_grow_spd(pipe, &spd))
324 return -ENOMEM;
325
Jens Axboecbb7e572006-04-11 14:57:50 +0200326 index = *ppos >> PAGE_CACHE_SHIFT;
Jens Axboe912d35f2006-04-26 10:59:21 +0200327 loff = *ppos & ~PAGE_CACHE_MASK;
Fengguang Wud8983912007-07-19 01:48:06 -0700328 req_pages = (len + loff + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
Jens Axboe35f3d142010-05-20 10:43:18 +0200329 nr_pages = min(req_pages, pipe->buffers);
Jens Axboe5274f052006-03-30 15:15:30 +0200330
331 /*
Jens Axboeeb207962006-04-27 11:05:22 +0200332 * Lookup the (hopefully) full range of pages we need.
333 */
Jens Axboe35f3d142010-05-20 10:43:18 +0200334 spd.nr_pages = find_get_pages_contig(mapping, index, nr_pages, spd.pages);
Fengguang Wu431a4822007-07-19 01:48:05 -0700335 index += spd.nr_pages;
Jens Axboeeb207962006-04-27 11:05:22 +0200336
337 /*
338 * If find_get_pages_contig() returned fewer pages than we needed,
Fengguang Wu431a4822007-07-19 01:48:05 -0700339 * readahead/allocate the rest and fill in the holes.
Jens Axboeeb207962006-04-27 11:05:22 +0200340 */
Fengguang Wu431a4822007-07-19 01:48:05 -0700341 if (spd.nr_pages < nr_pages)
Rusty Russellcf914a72007-07-19 01:48:08 -0700342 page_cache_sync_readahead(mapping, &in->f_ra, in,
343 index, req_pages - spd.nr_pages);
Fengguang Wu431a4822007-07-19 01:48:05 -0700344
Jens Axboe932cc6d2007-06-21 13:10:21 +0200345 error = 0;
Jens Axboeeb207962006-04-27 11:05:22 +0200346 while (spd.nr_pages < nr_pages) {
347 /*
348 * Page could be there, find_get_pages_contig() breaks on
349 * the first hole.
350 */
351 page = find_get_page(mapping, index);
352 if (!page) {
Jens Axboee27dedd2006-05-01 19:59:54 +0200353 /*
Jens Axboeeb207962006-04-27 11:05:22 +0200354 * page didn't exist, allocate one.
355 */
356 page = page_cache_alloc_cold(mapping);
357 if (!page)
358 break;
359
360 error = add_to_page_cache_lru(page, mapping, index,
Nick Piggin0ae0b5d2010-05-25 10:25:26 +0200361 GFP_KERNEL);
Jens Axboeeb207962006-04-27 11:05:22 +0200362 if (unlikely(error)) {
363 page_cache_release(page);
Jens Axboea0548872006-05-03 10:58:22 +0200364 if (error == -EEXIST)
365 continue;
Jens Axboeeb207962006-04-27 11:05:22 +0200366 break;
367 }
368 /*
369 * add_to_page_cache() locks the page, unlock it
370 * to avoid convoluting the logic below even more.
371 */
372 unlock_page(page);
373 }
374
Jens Axboe35f3d142010-05-20 10:43:18 +0200375 spd.pages[spd.nr_pages++] = page;
Jens Axboeeb207962006-04-27 11:05:22 +0200376 index++;
377 }
378
379 /*
380 * Now loop over the map and see if we need to start IO on any
381 * pages, fill in the partial map, etc.
382 */
383 index = *ppos >> PAGE_CACHE_SHIFT;
384 nr_pages = spd.nr_pages;
385 spd.nr_pages = 0;
386 for (page_nr = 0; page_nr < nr_pages; page_nr++) {
Jens Axboe82aa5d62006-04-20 13:05:48 +0200387 unsigned int this_len;
388
389 if (!len)
390 break;
391
392 /*
393 * this_len is the max we'll use from this page
394 */
Andrew Mortonba5f5d92006-04-25 15:33:34 +0200395 this_len = min_t(unsigned long, len, PAGE_CACHE_SIZE - loff);
Jens Axboe35f3d142010-05-20 10:43:18 +0200396 page = spd.pages[page_nr];
Jens Axboe7480a902006-04-11 13:52:47 +0200397
Fengguang Wua08a1662007-07-19 01:48:03 -0700398 if (PageReadahead(page))
Rusty Russellcf914a72007-07-19 01:48:08 -0700399 page_cache_async_readahead(mapping, &in->f_ra, in,
Fengguang Wud8983912007-07-19 01:48:06 -0700400 page, index, req_pages - page_nr);
Fengguang Wua08a1662007-07-19 01:48:03 -0700401
Jens Axboe7480a902006-04-11 13:52:47 +0200402 /*
403 * If the page isn't uptodate, we may need to start io on it
404 */
405 if (!PageUptodate(page)) {
Miklos Szeredi69650312010-08-03 12:48:50 +0200406 lock_page(page);
Jens Axboe7480a902006-04-11 13:52:47 +0200407
408 /*
Miklos Szeredi32502b82008-07-04 09:35:17 +0200409 * Page was truncated, or invalidated by the
410 * filesystem. Redo the find/create, but this time the
411 * page is kept locked, so there's no chance of another
412 * race with truncate/invalidate.
Jens Axboe7480a902006-04-11 13:52:47 +0200413 */
414 if (!page->mapping) {
415 unlock_page(page);
Miklos Szeredi32502b82008-07-04 09:35:17 +0200416 page = find_or_create_page(mapping, index,
417 mapping_gfp_mask(mapping));
418
419 if (!page) {
420 error = -ENOMEM;
421 break;
422 }
Jens Axboe35f3d142010-05-20 10:43:18 +0200423 page_cache_release(spd.pages[page_nr]);
424 spd.pages[page_nr] = page;
Jens Axboe7480a902006-04-11 13:52:47 +0200425 }
426 /*
427 * page was already under io and is now done, great
428 */
429 if (PageUptodate(page)) {
430 unlock_page(page);
431 goto fill_it;
432 }
433
Jens Axboe7480a902006-04-11 13:52:47 +0200434 /*
435 * need to read in the page
436 */
437 error = mapping->a_ops->readpage(in, page);
Jens Axboe7480a902006-04-11 13:52:47 +0200438 if (unlikely(error)) {
Jens Axboeeb207962006-04-27 11:05:22 +0200439 /*
440 * We really should re-lookup the page here,
441 * but it complicates things a lot. Instead
442 * lets just do what we already stored, and
443 * we'll get it the next time we are called.
444 */
Jens Axboe7480a902006-04-11 13:52:47 +0200445 if (error == AOP_TRUNCATED_PAGE)
Jens Axboeeb207962006-04-27 11:05:22 +0200446 error = 0;
447
Jens Axboe7480a902006-04-11 13:52:47 +0200448 break;
449 }
Jens Axboe620a3242007-06-07 09:39:42 +0200450 }
451fill_it:
452 /*
453 * i_size must be checked after PageUptodate.
454 */
455 isize = i_size_read(mapping->host);
456 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
457 if (unlikely(!isize || index > end_index))
458 break;
459
460 /*
461 * if this is the last page, see if we need to shrink
462 * the length and stop
463 */
464 if (end_index == index) {
465 unsigned int plen;
Jens Axboe91ad66e2006-04-19 15:55:10 +0200466
467 /*
Jens Axboe620a3242007-06-07 09:39:42 +0200468 * max good bytes in this page
Jens Axboe91ad66e2006-04-19 15:55:10 +0200469 */
Jens Axboe620a3242007-06-07 09:39:42 +0200470 plen = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
471 if (plen <= loff)
Jens Axboe91ad66e2006-04-19 15:55:10 +0200472 break;
Jens Axboe91ad66e2006-04-19 15:55:10 +0200473
474 /*
Jens Axboe620a3242007-06-07 09:39:42 +0200475 * force quit after adding this page
Jens Axboe91ad66e2006-04-19 15:55:10 +0200476 */
Jens Axboe620a3242007-06-07 09:39:42 +0200477 this_len = min(this_len, plen - loff);
478 len = this_len;
Jens Axboe7480a902006-04-11 13:52:47 +0200479 }
Jens Axboe620a3242007-06-07 09:39:42 +0200480
Jens Axboe35f3d142010-05-20 10:43:18 +0200481 spd.partial[page_nr].offset = loff;
482 spd.partial[page_nr].len = this_len;
Jens Axboe82aa5d62006-04-20 13:05:48 +0200483 len -= this_len;
Jens Axboe91ad66e2006-04-19 15:55:10 +0200484 loff = 0;
Jens Axboeeb207962006-04-27 11:05:22 +0200485 spd.nr_pages++;
486 index++;
Jens Axboe5274f052006-03-30 15:15:30 +0200487 }
488
Jens Axboeeb207962006-04-27 11:05:22 +0200489 /*
Hugh Dickins475ecad2007-06-07 09:36:00 +0200490 * Release any pages at the end, if we quit early. 'page_nr' is how far
Jens Axboeeb207962006-04-27 11:05:22 +0200491 * we got, 'nr_pages' is how many pages are in the map.
492 */
493 while (page_nr < nr_pages)
Jens Axboe35f3d142010-05-20 10:43:18 +0200494 page_cache_release(spd.pages[page_nr++]);
Fengguang Wuf4e6b492007-10-16 01:24:33 -0700495 in->f_ra.prev_pos = (loff_t)index << PAGE_CACHE_SHIFT;
Jens Axboeeb207962006-04-27 11:05:22 +0200496
Jens Axboe912d35f2006-04-26 10:59:21 +0200497 if (spd.nr_pages)
Jens Axboe35f3d142010-05-20 10:43:18 +0200498 error = splice_to_pipe(pipe, &spd);
Jens Axboe5274f052006-03-30 15:15:30 +0200499
Jens Axboe35f3d142010-05-20 10:43:18 +0200500 splice_shrink_spd(pipe, &spd);
Jens Axboe7480a902006-04-11 13:52:47 +0200501 return error;
Jens Axboe5274f052006-03-30 15:15:30 +0200502}
503
Jens Axboe83f91352006-04-02 23:05:09 +0200504/**
505 * generic_file_splice_read - splice data from file to a pipe
506 * @in: file to splice from
Jens Axboe932cc6d2007-06-21 13:10:21 +0200507 * @ppos: position in @in
Jens Axboe83f91352006-04-02 23:05:09 +0200508 * @pipe: pipe to splice to
509 * @len: number of bytes to splice
510 * @flags: splice modifier flags
511 *
Jens Axboe932cc6d2007-06-21 13:10:21 +0200512 * Description:
513 * Will read pages from given file and fill them into a pipe. Can be
514 * used as long as the address_space operations for the source implements
515 * a readpage() hook.
516 *
Jens Axboe83f91352006-04-02 23:05:09 +0200517 */
Jens Axboecbb7e572006-04-11 14:57:50 +0200518ssize_t generic_file_splice_read(struct file *in, loff_t *ppos,
519 struct pipe_inode_info *pipe, size_t len,
520 unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +0200521{
Jens Axboed366d3982007-06-01 14:54:11 +0200522 loff_t isize, left;
Jens Axboe8191ecd2008-04-10 08:24:25 +0200523 int ret;
Jens Axboed366d3982007-06-01 14:54:11 +0200524
525 isize = i_size_read(in->f_mapping->host);
526 if (unlikely(*ppos >= isize))
527 return 0;
528
529 left = isize - *ppos;
530 if (unlikely(left < len))
531 len = left;
Jens Axboe5274f052006-03-30 15:15:30 +0200532
Jens Axboe8191ecd2008-04-10 08:24:25 +0200533 ret = __generic_file_splice_read(in, ppos, pipe, len, flags);
Miklos Szeredi723590e2009-08-15 08:43:22 +0200534 if (ret > 0) {
Jens Axboecbb7e572006-04-11 14:57:50 +0200535 *ppos += ret;
Miklos Szeredi723590e2009-08-15 08:43:22 +0200536 file_accessed(in);
537 }
Jens Axboe5274f052006-03-30 15:15:30 +0200538
539 return ret;
540}
Jens Axboe059a8f32006-04-02 23:06:05 +0200541EXPORT_SYMBOL(generic_file_splice_read);
542
Miklos Szeredi68181732009-05-07 15:37:36 +0200543static const struct pipe_buf_operations default_pipe_buf_ops = {
544 .can_merge = 0,
545 .map = generic_pipe_buf_map,
546 .unmap = generic_pipe_buf_unmap,
547 .confirm = generic_pipe_buf_confirm,
548 .release = generic_pipe_buf_release,
549 .steal = generic_pipe_buf_steal,
550 .get = generic_pipe_buf_get,
551};
552
553static ssize_t kernel_readv(struct file *file, const struct iovec *vec,
554 unsigned long vlen, loff_t offset)
555{
556 mm_segment_t old_fs;
557 loff_t pos = offset;
558 ssize_t res;
559
560 old_fs = get_fs();
561 set_fs(get_ds());
562 /* The cast to a user pointer is valid due to the set_fs() */
563 res = vfs_readv(file, (const struct iovec __user *)vec, vlen, &pos);
564 set_fs(old_fs);
565
566 return res;
567}
568
Miklos Szeredib2858d72009-05-19 11:37:46 +0200569static ssize_t kernel_write(struct file *file, const char *buf, size_t count,
570 loff_t pos)
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +0200571{
572 mm_segment_t old_fs;
573 ssize_t res;
574
575 old_fs = get_fs();
576 set_fs(get_ds());
577 /* The cast to a user pointer is valid due to the set_fs() */
Miklos Szeredib2858d72009-05-19 11:37:46 +0200578 res = vfs_write(file, (const char __user *)buf, count, &pos);
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +0200579 set_fs(old_fs);
580
581 return res;
582}
583
Miklos Szeredi68181732009-05-07 15:37:36 +0200584ssize_t default_file_splice_read(struct file *in, loff_t *ppos,
585 struct pipe_inode_info *pipe, size_t len,
586 unsigned int flags)
587{
588 unsigned int nr_pages;
589 unsigned int nr_freed;
590 size_t offset;
Jens Axboe35f3d142010-05-20 10:43:18 +0200591 struct page *pages[PIPE_DEF_BUFFERS];
592 struct partial_page partial[PIPE_DEF_BUFFERS];
593 struct iovec *vec, __vec[PIPE_DEF_BUFFERS];
Miklos Szeredi68181732009-05-07 15:37:36 +0200594 ssize_t res;
595 size_t this_len;
596 int error;
597 int i;
598 struct splice_pipe_desc spd = {
599 .pages = pages,
600 .partial = partial,
601 .flags = flags,
602 .ops = &default_pipe_buf_ops,
603 .spd_release = spd_release_page,
604 };
605
Jens Axboe35f3d142010-05-20 10:43:18 +0200606 if (splice_grow_spd(pipe, &spd))
607 return -ENOMEM;
608
609 res = -ENOMEM;
610 vec = __vec;
611 if (pipe->buffers > PIPE_DEF_BUFFERS) {
612 vec = kmalloc(pipe->buffers * sizeof(struct iovec), GFP_KERNEL);
613 if (!vec)
614 goto shrink_ret;
615 }
616
Miklos Szeredi68181732009-05-07 15:37:36 +0200617 offset = *ppos & ~PAGE_CACHE_MASK;
618 nr_pages = (len + offset + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
619
Jens Axboe35f3d142010-05-20 10:43:18 +0200620 for (i = 0; i < nr_pages && i < pipe->buffers && len; i++) {
Miklos Szeredi68181732009-05-07 15:37:36 +0200621 struct page *page;
622
Jens Axboe4f231222009-05-13 08:35:35 +0200623 page = alloc_page(GFP_USER);
Miklos Szeredi68181732009-05-07 15:37:36 +0200624 error = -ENOMEM;
625 if (!page)
626 goto err;
627
628 this_len = min_t(size_t, len, PAGE_CACHE_SIZE - offset);
Jens Axboe4f231222009-05-13 08:35:35 +0200629 vec[i].iov_base = (void __user *) page_address(page);
Miklos Szeredi68181732009-05-07 15:37:36 +0200630 vec[i].iov_len = this_len;
Jens Axboe35f3d142010-05-20 10:43:18 +0200631 spd.pages[i] = page;
Miklos Szeredi68181732009-05-07 15:37:36 +0200632 spd.nr_pages++;
633 len -= this_len;
634 offset = 0;
635 }
636
637 res = kernel_readv(in, vec, spd.nr_pages, *ppos);
Andrew Morton77f6bf52009-05-14 09:49:44 +0200638 if (res < 0) {
639 error = res;
Miklos Szeredi68181732009-05-07 15:37:36 +0200640 goto err;
Andrew Morton77f6bf52009-05-14 09:49:44 +0200641 }
Miklos Szeredi68181732009-05-07 15:37:36 +0200642
643 error = 0;
644 if (!res)
645 goto err;
646
647 nr_freed = 0;
648 for (i = 0; i < spd.nr_pages; i++) {
Miklos Szeredi68181732009-05-07 15:37:36 +0200649 this_len = min_t(size_t, vec[i].iov_len, res);
Jens Axboe35f3d142010-05-20 10:43:18 +0200650 spd.partial[i].offset = 0;
651 spd.partial[i].len = this_len;
Miklos Szeredi68181732009-05-07 15:37:36 +0200652 if (!this_len) {
Jens Axboe35f3d142010-05-20 10:43:18 +0200653 __free_page(spd.pages[i]);
654 spd.pages[i] = NULL;
Miklos Szeredi68181732009-05-07 15:37:36 +0200655 nr_freed++;
656 }
657 res -= this_len;
658 }
659 spd.nr_pages -= nr_freed;
660
661 res = splice_to_pipe(pipe, &spd);
662 if (res > 0)
663 *ppos += res;
664
Jens Axboe35f3d142010-05-20 10:43:18 +0200665shrink_ret:
666 if (vec != __vec)
667 kfree(vec);
668 splice_shrink_spd(pipe, &spd);
Miklos Szeredi68181732009-05-07 15:37:36 +0200669 return res;
670
671err:
Jens Axboe4f231222009-05-13 08:35:35 +0200672 for (i = 0; i < spd.nr_pages; i++)
Jens Axboe35f3d142010-05-20 10:43:18 +0200673 __free_page(spd.pages[i]);
Jens Axboe4f231222009-05-13 08:35:35 +0200674
Jens Axboe35f3d142010-05-20 10:43:18 +0200675 res = error;
676 goto shrink_ret;
Miklos Szeredi68181732009-05-07 15:37:36 +0200677}
678EXPORT_SYMBOL(default_file_splice_read);
679
Jens Axboe5274f052006-03-30 15:15:30 +0200680/*
Jens Axboe4f6f0bd2006-04-02 23:04:46 +0200681 * Send 'sd->len' bytes to socket from 'sd->file' at position 'sd->pos'
Jens Axboe016b6612006-04-25 15:42:00 +0200682 * using sendpage(). Return the number of bytes sent.
Jens Axboe5274f052006-03-30 15:15:30 +0200683 */
Jens Axboe76ad4d12006-05-03 10:41:33 +0200684static int pipe_to_sendpage(struct pipe_inode_info *pipe,
Jens Axboe5274f052006-03-30 15:15:30 +0200685 struct pipe_buffer *buf, struct splice_desc *sd)
686{
Jens Axboe6a14b902007-06-14 13:08:55 +0200687 struct file *file = sd->u.file;
Jens Axboe5274f052006-03-30 15:15:30 +0200688 loff_t pos = sd->pos;
Michał Mirosława8adbe32010-12-17 08:56:44 +0100689 int more;
Jens Axboe5274f052006-03-30 15:15:30 +0200690
Michał Mirosława8adbe32010-12-17 08:56:44 +0100691 if (!likely(file->f_op && file->f_op->sendpage))
692 return -EINVAL;
Jens Axboe5274f052006-03-30 15:15:30 +0200693
Michał Mirosława8adbe32010-12-17 08:56:44 +0100694 more = (sd->flags & SPLICE_F_MORE) || sd->len < sd->total_len;
695 return file->f_op->sendpage(file, buf->page, buf->offset,
696 sd->len, &pos, more);
Jens Axboe5274f052006-03-30 15:15:30 +0200697}
698
699/*
700 * This is a little more tricky than the file -> pipe splicing. There are
701 * basically three cases:
702 *
703 * - Destination page already exists in the address space and there
704 * are users of it. For that case we have no other option that
705 * copying the data. Tough luck.
706 * - Destination page already exists in the address space, but there
707 * are no users of it. Make sure it's uptodate, then drop it. Fall
708 * through to last case.
709 * - Destination page does not exist, we can add the pipe page to
710 * the page cache and avoid the copy.
711 *
Jens Axboe83f91352006-04-02 23:05:09 +0200712 * If asked to move pages to the output file (SPLICE_F_MOVE is set in
713 * sd->flags), we attempt to migrate pages from the pipe to the output
714 * file address space page cache. This is possible if no one else has
715 * the pipe page referenced outside of the pipe and page cache. If
716 * SPLICE_F_MOVE isn't set, or we cannot move the page, we simply create
717 * a new page in the output file page cache and fill/dirty that.
Jens Axboe5274f052006-03-30 15:15:30 +0200718 */
Miklos Szeredi328eaab2009-04-14 19:48:39 +0200719int pipe_to_file(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
720 struct splice_desc *sd)
Jens Axboe5274f052006-03-30 15:15:30 +0200721{
Jens Axboe6a14b902007-06-14 13:08:55 +0200722 struct file *file = sd->u.file;
Jens Axboe5274f052006-03-30 15:15:30 +0200723 struct address_space *mapping = file->f_mapping;
Jens Axboe016b6612006-04-25 15:42:00 +0200724 unsigned int offset, this_len;
Jens Axboe5274f052006-03-30 15:15:30 +0200725 struct page *page;
Nick Pigginafddba42007-10-16 01:25:01 -0700726 void *fsdata;
Jens Axboe3e7ee3e2006-04-02 23:11:04 +0200727 int ret;
Jens Axboe5274f052006-03-30 15:15:30 +0200728
Jens Axboe5274f052006-03-30 15:15:30 +0200729 offset = sd->pos & ~PAGE_CACHE_MASK;
730
Jens Axboe016b6612006-04-25 15:42:00 +0200731 this_len = sd->len;
732 if (this_len + offset > PAGE_CACHE_SIZE)
733 this_len = PAGE_CACHE_SIZE - offset;
734
Nick Pigginafddba42007-10-16 01:25:01 -0700735 ret = pagecache_write_begin(file, mapping, sd->pos, this_len,
736 AOP_FLAG_UNINTERRUPTIBLE, &page, &fsdata);
737 if (unlikely(ret))
738 goto out;
Jens Axboe5274f052006-03-30 15:15:30 +0200739
Jens Axboe0568b402006-05-01 19:50:48 +0200740 if (buf->page != page) {
Jens Axboef84d7512006-05-01 19:59:03 +0200741 /*
742 * Careful, ->map() uses KM_USER0!
743 */
Jens Axboe76ad4d12006-05-03 10:41:33 +0200744 char *src = buf->ops->map(pipe, buf, 1);
Jens Axboef84d7512006-05-01 19:59:03 +0200745 char *dst = kmap_atomic(page, KM_USER1);
Jens Axboe5abc97a2006-03-30 15:16:46 +0200746
Jens Axboe016b6612006-04-25 15:42:00 +0200747 memcpy(dst + offset, src + buf->offset, this_len);
Jens Axboe5abc97a2006-03-30 15:16:46 +0200748 flush_dcache_page(page);
Jens Axboef84d7512006-05-01 19:59:03 +0200749 kunmap_atomic(dst, KM_USER1);
Jens Axboe76ad4d12006-05-03 10:41:33 +0200750 buf->ops->unmap(pipe, buf, src);
Jens Axboe5abc97a2006-03-30 15:16:46 +0200751 }
Nick Pigginafddba42007-10-16 01:25:01 -0700752 ret = pagecache_write_end(file, mapping, sd->pos, this_len, this_len,
753 page, fsdata);
Jens Axboe5274f052006-03-30 15:15:30 +0200754out:
Jens Axboe5274f052006-03-30 15:15:30 +0200755 return ret;
756}
Miklos Szeredi328eaab2009-04-14 19:48:39 +0200757EXPORT_SYMBOL(pipe_to_file);
Jens Axboe5274f052006-03-30 15:15:30 +0200758
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200759static void wakeup_pipe_writers(struct pipe_inode_info *pipe)
760{
761 smp_mb();
762 if (waitqueue_active(&pipe->wait))
763 wake_up_interruptible(&pipe->wait);
764 kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
765}
766
767/**
768 * splice_from_pipe_feed - feed available data from a pipe to a file
769 * @pipe: pipe to splice from
770 * @sd: information to @actor
771 * @actor: handler that splices the data
772 *
773 * Description:
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200774 * This function loops over the pipe and calls @actor to do the
775 * actual moving of a single struct pipe_buffer to the desired
776 * destination. It returns when there's no more buffers left in
777 * the pipe or if the requested number of bytes (@sd->total_len)
778 * have been copied. It returns a positive number (one) if the
779 * pipe needs to be filled with more data, zero if the required
780 * number of bytes have been copied and -errno on error.
781 *
782 * This, together with splice_from_pipe_{begin,end,next}, may be
783 * used to implement the functionality of __splice_from_pipe() when
784 * locking is required around copying the pipe buffers to the
785 * destination.
786 */
787int splice_from_pipe_feed(struct pipe_inode_info *pipe, struct splice_desc *sd,
788 splice_actor *actor)
789{
790 int ret;
791
792 while (pipe->nrbufs) {
793 struct pipe_buffer *buf = pipe->bufs + pipe->curbuf;
794 const struct pipe_buf_operations *ops = buf->ops;
795
796 sd->len = buf->len;
797 if (sd->len > sd->total_len)
798 sd->len = sd->total_len;
799
Michał Mirosława8adbe32010-12-17 08:56:44 +0100800 ret = buf->ops->confirm(pipe, buf);
801 if (unlikely(ret)) {
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200802 if (ret == -ENODATA)
803 ret = 0;
804 return ret;
805 }
Michał Mirosława8adbe32010-12-17 08:56:44 +0100806
807 ret = actor(pipe, buf, sd);
808 if (ret <= 0)
809 return ret;
810
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200811 buf->offset += ret;
812 buf->len -= ret;
813
814 sd->num_spliced += ret;
815 sd->len -= ret;
816 sd->pos += ret;
817 sd->total_len -= ret;
818
819 if (!buf->len) {
820 buf->ops = NULL;
821 ops->release(pipe, buf);
Jens Axboe35f3d142010-05-20 10:43:18 +0200822 pipe->curbuf = (pipe->curbuf + 1) & (pipe->buffers - 1);
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200823 pipe->nrbufs--;
824 if (pipe->inode)
825 sd->need_wakeup = true;
826 }
827
828 if (!sd->total_len)
829 return 0;
830 }
831
832 return 1;
833}
834EXPORT_SYMBOL(splice_from_pipe_feed);
835
836/**
837 * splice_from_pipe_next - wait for some data to splice from
838 * @pipe: pipe to splice from
839 * @sd: information about the splice operation
840 *
841 * Description:
842 * This function will wait for some data and return a positive
843 * value (one) if pipe buffers are available. It will return zero
844 * or -errno if no more data needs to be spliced.
845 */
846int splice_from_pipe_next(struct pipe_inode_info *pipe, struct splice_desc *sd)
847{
848 while (!pipe->nrbufs) {
849 if (!pipe->writers)
850 return 0;
851
852 if (!pipe->waiting_writers && sd->num_spliced)
853 return 0;
854
855 if (sd->flags & SPLICE_F_NONBLOCK)
856 return -EAGAIN;
857
858 if (signal_pending(current))
859 return -ERESTARTSYS;
860
861 if (sd->need_wakeup) {
862 wakeup_pipe_writers(pipe);
863 sd->need_wakeup = false;
864 }
865
866 pipe_wait(pipe);
867 }
868
869 return 1;
870}
871EXPORT_SYMBOL(splice_from_pipe_next);
872
873/**
874 * splice_from_pipe_begin - start splicing from pipe
Randy Dunlapb80901b2009-04-16 19:09:55 -0700875 * @sd: information about the splice operation
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200876 *
877 * Description:
878 * This function should be called before a loop containing
879 * splice_from_pipe_next() and splice_from_pipe_feed() to
880 * initialize the necessary fields of @sd.
881 */
882void splice_from_pipe_begin(struct splice_desc *sd)
883{
884 sd->num_spliced = 0;
885 sd->need_wakeup = false;
886}
887EXPORT_SYMBOL(splice_from_pipe_begin);
888
889/**
890 * splice_from_pipe_end - finish splicing from pipe
891 * @pipe: pipe to splice from
892 * @sd: information about the splice operation
893 *
894 * Description:
895 * This function will wake up pipe writers if necessary. It should
896 * be called after a loop containing splice_from_pipe_next() and
897 * splice_from_pipe_feed().
898 */
899void splice_from_pipe_end(struct pipe_inode_info *pipe, struct splice_desc *sd)
900{
901 if (sd->need_wakeup)
902 wakeup_pipe_writers(pipe);
903}
904EXPORT_SYMBOL(splice_from_pipe_end);
905
Jens Axboe932cc6d2007-06-21 13:10:21 +0200906/**
907 * __splice_from_pipe - splice data from a pipe to given actor
908 * @pipe: pipe to splice from
909 * @sd: information to @actor
910 * @actor: handler that splices the data
911 *
912 * Description:
913 * This function does little more than loop over the pipe and call
914 * @actor to do the actual moving of a single struct pipe_buffer to
915 * the desired destination. See pipe_to_file, pipe_to_sendpage, or
916 * pipe_to_user.
917 *
Jens Axboe83f91352006-04-02 23:05:09 +0200918 */
Jens Axboec66ab6f2007-06-12 21:17:17 +0200919ssize_t __splice_from_pipe(struct pipe_inode_info *pipe, struct splice_desc *sd,
920 splice_actor *actor)
Jens Axboe5274f052006-03-30 15:15:30 +0200921{
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200922 int ret;
Jens Axboe5274f052006-03-30 15:15:30 +0200923
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200924 splice_from_pipe_begin(sd);
925 do {
926 ret = splice_from_pipe_next(pipe, sd);
927 if (ret > 0)
928 ret = splice_from_pipe_feed(pipe, sd, actor);
929 } while (ret > 0);
930 splice_from_pipe_end(pipe, sd);
Jens Axboe5274f052006-03-30 15:15:30 +0200931
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200932 return sd->num_spliced ? sd->num_spliced : ret;
Jens Axboe5274f052006-03-30 15:15:30 +0200933}
Mark Fasheh40bee442007-03-21 13:11:02 +0100934EXPORT_SYMBOL(__splice_from_pipe);
Jens Axboe5274f052006-03-30 15:15:30 +0200935
Jens Axboe932cc6d2007-06-21 13:10:21 +0200936/**
937 * splice_from_pipe - splice data from a pipe to a file
938 * @pipe: pipe to splice from
939 * @out: file to splice to
940 * @ppos: position in @out
941 * @len: how many bytes to splice
942 * @flags: splice modifier flags
943 * @actor: handler that splices the data
944 *
945 * Description:
Miklos Szeredi29339702009-04-14 19:48:37 +0200946 * See __splice_from_pipe. This function locks the pipe inode,
Jens Axboe932cc6d2007-06-21 13:10:21 +0200947 * otherwise it's identical to __splice_from_pipe().
948 *
949 */
Mark Fasheh6da61802006-10-17 18:43:07 +0200950ssize_t splice_from_pipe(struct pipe_inode_info *pipe, struct file *out,
951 loff_t *ppos, size_t len, unsigned int flags,
952 splice_actor *actor)
953{
954 ssize_t ret;
Jens Axboec66ab6f2007-06-12 21:17:17 +0200955 struct splice_desc sd = {
956 .total_len = len,
957 .flags = flags,
958 .pos = *ppos,
Jens Axboe6a14b902007-06-14 13:08:55 +0200959 .u.file = out,
Jens Axboec66ab6f2007-06-12 21:17:17 +0200960 };
Mark Fasheh6da61802006-10-17 18:43:07 +0200961
Miklos Szeredi61e0d472009-04-14 19:48:41 +0200962 pipe_lock(pipe);
Jens Axboec66ab6f2007-06-12 21:17:17 +0200963 ret = __splice_from_pipe(pipe, &sd, actor);
Miklos Szeredi61e0d472009-04-14 19:48:41 +0200964 pipe_unlock(pipe);
Mark Fasheh6da61802006-10-17 18:43:07 +0200965
966 return ret;
967}
968
969/**
Jens Axboe83f91352006-04-02 23:05:09 +0200970 * generic_file_splice_write - splice data from a pipe to a file
Ingo Molnar3a326a22006-04-10 15:18:35 +0200971 * @pipe: pipe info
Jens Axboe83f91352006-04-02 23:05:09 +0200972 * @out: file to write to
Jens Axboe932cc6d2007-06-21 13:10:21 +0200973 * @ppos: position in @out
Jens Axboe83f91352006-04-02 23:05:09 +0200974 * @len: number of bytes to splice
975 * @flags: splice modifier flags
976 *
Jens Axboe932cc6d2007-06-21 13:10:21 +0200977 * Description:
978 * Will either move or copy pages (determined by @flags options) from
979 * the given pipe inode to the given file.
Jens Axboe83f91352006-04-02 23:05:09 +0200980 *
981 */
Ingo Molnar3a326a22006-04-10 15:18:35 +0200982ssize_t
983generic_file_splice_write(struct pipe_inode_info *pipe, struct file *out,
Jens Axboecbb7e572006-04-11 14:57:50 +0200984 loff_t *ppos, size_t len, unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +0200985{
Jens Axboe4f6f0bd2006-04-02 23:04:46 +0200986 struct address_space *mapping = out->f_mapping;
Jens Axboe8c34e2d2006-10-17 19:43:22 +0200987 struct inode *inode = mapping->host;
Miklos Szeredi7f3d4ee2008-05-07 09:22:39 +0200988 struct splice_desc sd = {
989 .total_len = len,
990 .flags = flags,
991 .pos = *ppos,
992 .u.file = out,
993 };
Ingo Molnar3a326a22006-04-10 15:18:35 +0200994 ssize_t ret;
Jens Axboe8c34e2d2006-10-17 19:43:22 +0200995
Miklos Szeredi61e0d472009-04-14 19:48:41 +0200996 pipe_lock(pipe);
Miklos Szeredieb443e52009-04-14 19:48:38 +0200997
998 splice_from_pipe_begin(&sd);
999 do {
1000 ret = splice_from_pipe_next(pipe, &sd);
1001 if (ret <= 0)
1002 break;
1003
1004 mutex_lock_nested(&inode->i_mutex, I_MUTEX_CHILD);
1005 ret = file_remove_suid(out);
Miklos Szeredi723590e2009-08-15 08:43:22 +02001006 if (!ret) {
1007 file_update_time(out);
Miklos Szeredieb443e52009-04-14 19:48:38 +02001008 ret = splice_from_pipe_feed(pipe, &sd, pipe_to_file);
Miklos Szeredi723590e2009-08-15 08:43:22 +02001009 }
Miklos Szeredieb443e52009-04-14 19:48:38 +02001010 mutex_unlock(&inode->i_mutex);
1011 } while (ret > 0);
1012 splice_from_pipe_end(pipe, &sd);
1013
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001014 pipe_unlock(pipe);
Miklos Szeredieb443e52009-04-14 19:48:38 +02001015
1016 if (sd.num_spliced)
1017 ret = sd.num_spliced;
1018
Jens Axboea4514eb2006-04-19 15:57:05 +02001019 if (ret > 0) {
Jens Axboe17ee4f42007-06-15 13:10:37 +02001020 unsigned long nr_pages;
Jan Kara148f9482009-08-17 19:52:36 +02001021 int err;
Jens Axboe17ee4f42007-06-15 13:10:37 +02001022
Jens Axboe17ee4f42007-06-15 13:10:37 +02001023 nr_pages = (ret + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
Jens Axboe4f6f0bd2006-04-02 23:04:46 +02001024
Jan Kara148f9482009-08-17 19:52:36 +02001025 err = generic_write_sync(out, *ppos, ret);
1026 if (err)
1027 ret = err;
1028 else
1029 *ppos += ret;
Jens Axboe17ee4f42007-06-15 13:10:37 +02001030 balance_dirty_pages_ratelimited_nr(mapping, nr_pages);
Jens Axboe4f6f0bd2006-04-02 23:04:46 +02001031 }
1032
1033 return ret;
Jens Axboe5274f052006-03-30 15:15:30 +02001034}
1035
Jens Axboe059a8f32006-04-02 23:06:05 +02001036EXPORT_SYMBOL(generic_file_splice_write);
1037
Miklos Szeredib2858d72009-05-19 11:37:46 +02001038static int write_pipe_buf(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
1039 struct splice_desc *sd)
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +02001040{
Miklos Szeredib2858d72009-05-19 11:37:46 +02001041 int ret;
1042 void *data;
1043
Miklos Szeredib2858d72009-05-19 11:37:46 +02001044 data = buf->ops->map(pipe, buf, 0);
1045 ret = kernel_write(sd->u.file, data + buf->offset, sd->len, sd->pos);
1046 buf->ops->unmap(pipe, buf, data);
1047
1048 return ret;
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +02001049}
1050
1051static ssize_t default_file_splice_write(struct pipe_inode_info *pipe,
1052 struct file *out, loff_t *ppos,
1053 size_t len, unsigned int flags)
1054{
Miklos Szeredib2858d72009-05-19 11:37:46 +02001055 ssize_t ret;
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +02001056
Miklos Szeredib2858d72009-05-19 11:37:46 +02001057 ret = splice_from_pipe(pipe, out, ppos, len, flags, write_pipe_buf);
1058 if (ret > 0)
1059 *ppos += ret;
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +02001060
Miklos Szeredib2858d72009-05-19 11:37:46 +02001061 return ret;
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +02001062}
1063
Jens Axboe83f91352006-04-02 23:05:09 +02001064/**
1065 * generic_splice_sendpage - splice data from a pipe to a socket
Jens Axboe932cc6d2007-06-21 13:10:21 +02001066 * @pipe: pipe to splice from
Jens Axboe83f91352006-04-02 23:05:09 +02001067 * @out: socket to write to
Jens Axboe932cc6d2007-06-21 13:10:21 +02001068 * @ppos: position in @out
Jens Axboe83f91352006-04-02 23:05:09 +02001069 * @len: number of bytes to splice
1070 * @flags: splice modifier flags
1071 *
Jens Axboe932cc6d2007-06-21 13:10:21 +02001072 * Description:
1073 * Will send @len bytes from the pipe to a network socket. No data copying
1074 * is involved.
Jens Axboe83f91352006-04-02 23:05:09 +02001075 *
1076 */
Ingo Molnar3a326a22006-04-10 15:18:35 +02001077ssize_t generic_splice_sendpage(struct pipe_inode_info *pipe, struct file *out,
Jens Axboecbb7e572006-04-11 14:57:50 +02001078 loff_t *ppos, size_t len, unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +02001079{
Jens Axboe00522fb2006-04-26 14:39:29 +02001080 return splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_sendpage);
Jens Axboe5274f052006-03-30 15:15:30 +02001081}
1082
Jens Axboe059a8f32006-04-02 23:06:05 +02001083EXPORT_SYMBOL(generic_splice_sendpage);
Jeff Garzika0f06782006-03-30 23:06:13 -05001084
Jens Axboe83f91352006-04-02 23:05:09 +02001085/*
1086 * Attempt to initiate a splice from pipe to file.
1087 */
Ingo Molnar3a326a22006-04-10 15:18:35 +02001088static long do_splice_from(struct pipe_inode_info *pipe, struct file *out,
Jens Axboecbb7e572006-04-11 14:57:50 +02001089 loff_t *ppos, size_t len, unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +02001090{
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +02001091 ssize_t (*splice_write)(struct pipe_inode_info *, struct file *,
1092 loff_t *, size_t, unsigned int);
Jens Axboe5274f052006-03-30 15:15:30 +02001093 int ret;
1094
Jens Axboe49570e92006-04-11 13:56:09 +02001095 if (unlikely(!(out->f_mode & FMODE_WRITE)))
Jens Axboe5274f052006-03-30 15:15:30 +02001096 return -EBADF;
1097
Linus Torvaldsefc968d2008-10-09 14:04:54 -07001098 if (unlikely(out->f_flags & O_APPEND))
1099 return -EINVAL;
1100
Jens Axboecbb7e572006-04-11 14:57:50 +02001101 ret = rw_verify_area(WRITE, out, ppos, len);
Jens Axboe5274f052006-03-30 15:15:30 +02001102 if (unlikely(ret < 0))
1103 return ret;
1104
Changli Gaocc56f7d2009-11-04 09:09:52 +01001105 if (out->f_op && out->f_op->splice_write)
1106 splice_write = out->f_op->splice_write;
1107 else
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +02001108 splice_write = default_file_splice_write;
1109
1110 return splice_write(pipe, out, ppos, len, flags);
Jens Axboe5274f052006-03-30 15:15:30 +02001111}
1112
Jens Axboe83f91352006-04-02 23:05:09 +02001113/*
1114 * Attempt to initiate a splice from a file to a pipe.
1115 */
Jens Axboecbb7e572006-04-11 14:57:50 +02001116static long do_splice_to(struct file *in, loff_t *ppos,
1117 struct pipe_inode_info *pipe, size_t len,
1118 unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +02001119{
Miklos Szeredi68181732009-05-07 15:37:36 +02001120 ssize_t (*splice_read)(struct file *, loff_t *,
1121 struct pipe_inode_info *, size_t, unsigned int);
Jens Axboe5274f052006-03-30 15:15:30 +02001122 int ret;
1123
Jens Axboe49570e92006-04-11 13:56:09 +02001124 if (unlikely(!(in->f_mode & FMODE_READ)))
Jens Axboe5274f052006-03-30 15:15:30 +02001125 return -EBADF;
1126
Jens Axboecbb7e572006-04-11 14:57:50 +02001127 ret = rw_verify_area(READ, in, ppos, len);
Jens Axboe5274f052006-03-30 15:15:30 +02001128 if (unlikely(ret < 0))
1129 return ret;
1130
Changli Gaocc56f7d2009-11-04 09:09:52 +01001131 if (in->f_op && in->f_op->splice_read)
1132 splice_read = in->f_op->splice_read;
1133 else
Miklos Szeredi68181732009-05-07 15:37:36 +02001134 splice_read = default_file_splice_read;
1135
1136 return splice_read(in, ppos, pipe, len, flags);
Jens Axboe5274f052006-03-30 15:15:30 +02001137}
1138
Jens Axboe932cc6d2007-06-21 13:10:21 +02001139/**
1140 * splice_direct_to_actor - splices data directly between two non-pipes
1141 * @in: file to splice from
1142 * @sd: actor information on where to splice to
1143 * @actor: handles the data splicing
1144 *
1145 * Description:
1146 * This is a special case helper to splice directly between two
1147 * points, without requiring an explicit pipe. Internally an allocated
Randy Dunlap79685b82007-07-27 08:08:51 +02001148 * pipe is cached in the process, and reused during the lifetime of
Jens Axboe932cc6d2007-06-21 13:10:21 +02001149 * that process.
1150 *
Jens Axboec66ab6f2007-06-12 21:17:17 +02001151 */
1152ssize_t splice_direct_to_actor(struct file *in, struct splice_desc *sd,
1153 splice_direct_actor *actor)
Jens Axboeb92ce552006-04-11 13:52:07 +02001154{
1155 struct pipe_inode_info *pipe;
1156 long ret, bytes;
1157 umode_t i_mode;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001158 size_t len;
1159 int i, flags;
Jens Axboeb92ce552006-04-11 13:52:07 +02001160
1161 /*
1162 * We require the input being a regular file, as we don't want to
1163 * randomly drop data for eg socket -> socket splicing. Use the
1164 * piped splicing for that!
1165 */
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001166 i_mode = in->f_path.dentry->d_inode->i_mode;
Jens Axboeb92ce552006-04-11 13:52:07 +02001167 if (unlikely(!S_ISREG(i_mode) && !S_ISBLK(i_mode)))
1168 return -EINVAL;
1169
1170 /*
1171 * neither in nor out is a pipe, setup an internal pipe attached to
1172 * 'out' and transfer the wanted data from 'in' to 'out' through that
1173 */
1174 pipe = current->splice_pipe;
Jens Axboe49570e92006-04-11 13:56:09 +02001175 if (unlikely(!pipe)) {
Jens Axboeb92ce552006-04-11 13:52:07 +02001176 pipe = alloc_pipe_info(NULL);
1177 if (!pipe)
1178 return -ENOMEM;
1179
1180 /*
1181 * We don't have an immediate reader, but we'll read the stuff
Jens Axboe00522fb2006-04-26 14:39:29 +02001182 * out of the pipe right after the splice_to_pipe(). So set
Jens Axboeb92ce552006-04-11 13:52:07 +02001183 * PIPE_READERS appropriately.
1184 */
1185 pipe->readers = 1;
1186
1187 current->splice_pipe = pipe;
1188 }
1189
1190 /*
Ingo Molnar73d62d82006-04-11 13:57:21 +02001191 * Do the splice.
Jens Axboeb92ce552006-04-11 13:52:07 +02001192 */
1193 ret = 0;
1194 bytes = 0;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001195 len = sd->total_len;
1196 flags = sd->flags;
1197
1198 /*
1199 * Don't block on output, we have to drain the direct pipe.
1200 */
1201 sd->flags &= ~SPLICE_F_NONBLOCK;
Jens Axboeb92ce552006-04-11 13:52:07 +02001202
1203 while (len) {
Jens Axboe51a92c02007-07-13 14:11:43 +02001204 size_t read_len;
Tom Zanussia82c53a2008-05-09 13:28:36 +02001205 loff_t pos = sd->pos, prev_pos = pos;
Jens Axboeb92ce552006-04-11 13:52:07 +02001206
Jens Axboebcd4f3a2007-07-16 14:41:49 +02001207 ret = do_splice_to(in, &pos, pipe, len, flags);
Jens Axboe51a92c02007-07-13 14:11:43 +02001208 if (unlikely(ret <= 0))
Jens Axboeb92ce552006-04-11 13:52:07 +02001209 goto out_release;
1210
1211 read_len = ret;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001212 sd->total_len = read_len;
Jens Axboeb92ce552006-04-11 13:52:07 +02001213
1214 /*
1215 * NOTE: nonblocking mode only applies to the input. We
1216 * must not do the output in nonblocking mode as then we
1217 * could get stuck data in the internal pipe:
1218 */
Jens Axboec66ab6f2007-06-12 21:17:17 +02001219 ret = actor(pipe, sd);
Tom Zanussia82c53a2008-05-09 13:28:36 +02001220 if (unlikely(ret <= 0)) {
1221 sd->pos = prev_pos;
Jens Axboeb92ce552006-04-11 13:52:07 +02001222 goto out_release;
Tom Zanussia82c53a2008-05-09 13:28:36 +02001223 }
Jens Axboeb92ce552006-04-11 13:52:07 +02001224
1225 bytes += ret;
1226 len -= ret;
Jens Axboebcd4f3a2007-07-16 14:41:49 +02001227 sd->pos = pos;
Jens Axboeb92ce552006-04-11 13:52:07 +02001228
Tom Zanussia82c53a2008-05-09 13:28:36 +02001229 if (ret < read_len) {
1230 sd->pos = prev_pos + ret;
Jens Axboe51a92c02007-07-13 14:11:43 +02001231 goto out_release;
Tom Zanussia82c53a2008-05-09 13:28:36 +02001232 }
Jens Axboeb92ce552006-04-11 13:52:07 +02001233 }
1234
Jens Axboe9e971982008-01-29 21:05:57 +01001235done:
Jens Axboeb92ce552006-04-11 13:52:07 +02001236 pipe->nrbufs = pipe->curbuf = 0;
Jens Axboe80848702008-01-30 12:24:48 +01001237 file_accessed(in);
Jens Axboeb92ce552006-04-11 13:52:07 +02001238 return bytes;
1239
1240out_release:
1241 /*
1242 * If we did an incomplete transfer we must release
1243 * the pipe buffers in question:
1244 */
Jens Axboe35f3d142010-05-20 10:43:18 +02001245 for (i = 0; i < pipe->buffers; i++) {
Jens Axboeb92ce552006-04-11 13:52:07 +02001246 struct pipe_buffer *buf = pipe->bufs + i;
1247
1248 if (buf->ops) {
1249 buf->ops->release(pipe, buf);
1250 buf->ops = NULL;
1251 }
1252 }
Jens Axboeb92ce552006-04-11 13:52:07 +02001253
Jens Axboe9e971982008-01-29 21:05:57 +01001254 if (!bytes)
1255 bytes = ret;
Jens Axboeb92ce552006-04-11 13:52:07 +02001256
Jens Axboe9e971982008-01-29 21:05:57 +01001257 goto done;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001258}
1259EXPORT_SYMBOL(splice_direct_to_actor);
1260
1261static int direct_splice_actor(struct pipe_inode_info *pipe,
1262 struct splice_desc *sd)
1263{
Jens Axboe6a14b902007-06-14 13:08:55 +02001264 struct file *file = sd->u.file;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001265
Changli Gao2cb4b052010-06-29 13:09:18 +02001266 return do_splice_from(pipe, file, &file->f_pos, sd->total_len,
1267 sd->flags);
Jens Axboec66ab6f2007-06-12 21:17:17 +02001268}
1269
Jens Axboe932cc6d2007-06-21 13:10:21 +02001270/**
1271 * do_splice_direct - splices data directly between two files
1272 * @in: file to splice from
1273 * @ppos: input file offset
1274 * @out: file to splice to
1275 * @len: number of bytes to splice
1276 * @flags: splice modifier flags
1277 *
1278 * Description:
1279 * For use by do_sendfile(). splice can easily emulate sendfile, but
1280 * doing it in the application would incur an extra system call
1281 * (splice in + splice out, as compared to just sendfile()). So this helper
1282 * can splice directly through a process-private pipe.
1283 *
1284 */
Jens Axboec66ab6f2007-06-12 21:17:17 +02001285long do_splice_direct(struct file *in, loff_t *ppos, struct file *out,
1286 size_t len, unsigned int flags)
1287{
1288 struct splice_desc sd = {
1289 .len = len,
1290 .total_len = len,
1291 .flags = flags,
1292 .pos = *ppos,
Jens Axboe6a14b902007-06-14 13:08:55 +02001293 .u.file = out,
Jens Axboec66ab6f2007-06-12 21:17:17 +02001294 };
Jens Axboe51a92c02007-07-13 14:11:43 +02001295 long ret;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001296
1297 ret = splice_direct_to_actor(in, &sd, direct_splice_actor);
Jens Axboe51a92c02007-07-13 14:11:43 +02001298 if (ret > 0)
Tom Zanussia82c53a2008-05-09 13:28:36 +02001299 *ppos = sd.pos;
Jens Axboe51a92c02007-07-13 14:11:43 +02001300
Jens Axboec66ab6f2007-06-12 21:17:17 +02001301 return ret;
Jens Axboeb92ce552006-04-11 13:52:07 +02001302}
1303
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001304static int splice_pipe_to_pipe(struct pipe_inode_info *ipipe,
1305 struct pipe_inode_info *opipe,
1306 size_t len, unsigned int flags);
Jens Axboeddac0d32006-11-04 12:49:32 +01001307
1308/*
Jens Axboe83f91352006-04-02 23:05:09 +02001309 * Determine where to splice to/from.
1310 */
Ingo Molnar529565d2006-04-10 15:18:58 +02001311static long do_splice(struct file *in, loff_t __user *off_in,
1312 struct file *out, loff_t __user *off_out,
1313 size_t len, unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +02001314{
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001315 struct pipe_inode_info *ipipe;
1316 struct pipe_inode_info *opipe;
Jens Axboecbb7e572006-04-11 14:57:50 +02001317 loff_t offset, *off;
Jens Axboea4514eb2006-04-19 15:57:05 +02001318 long ret;
Jens Axboe5274f052006-03-30 15:15:30 +02001319
Linus Torvalds71993e62010-11-28 13:56:09 -08001320 ipipe = get_pipe_info(in);
1321 opipe = get_pipe_info(out);
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001322
1323 if (ipipe && opipe) {
1324 if (off_in || off_out)
1325 return -ESPIPE;
1326
1327 if (!(in->f_mode & FMODE_READ))
1328 return -EBADF;
1329
1330 if (!(out->f_mode & FMODE_WRITE))
1331 return -EBADF;
1332
1333 /* Splicing to self would be fun, but... */
1334 if (ipipe == opipe)
1335 return -EINVAL;
1336
1337 return splice_pipe_to_pipe(ipipe, opipe, len, flags);
1338 }
1339
1340 if (ipipe) {
Ingo Molnar529565d2006-04-10 15:18:58 +02001341 if (off_in)
1342 return -ESPIPE;
Jens Axboeb92ce552006-04-11 13:52:07 +02001343 if (off_out) {
Changli Gao19c9a492010-06-29 13:10:36 +02001344 if (!(out->f_mode & FMODE_PWRITE))
Jens Axboeb92ce552006-04-11 13:52:07 +02001345 return -EINVAL;
Jens Axboecbb7e572006-04-11 14:57:50 +02001346 if (copy_from_user(&offset, off_out, sizeof(loff_t)))
Jens Axboeb92ce552006-04-11 13:52:07 +02001347 return -EFAULT;
Jens Axboecbb7e572006-04-11 14:57:50 +02001348 off = &offset;
1349 } else
1350 off = &out->f_pos;
Ingo Molnar529565d2006-04-10 15:18:58 +02001351
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001352 ret = do_splice_from(ipipe, out, off, len, flags);
Jens Axboea4514eb2006-04-19 15:57:05 +02001353
1354 if (off_out && copy_to_user(off_out, off, sizeof(loff_t)))
1355 ret = -EFAULT;
1356
1357 return ret;
Ingo Molnar529565d2006-04-10 15:18:58 +02001358 }
Jens Axboe5274f052006-03-30 15:15:30 +02001359
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001360 if (opipe) {
Ingo Molnar529565d2006-04-10 15:18:58 +02001361 if (off_out)
1362 return -ESPIPE;
Jens Axboeb92ce552006-04-11 13:52:07 +02001363 if (off_in) {
Changli Gao19c9a492010-06-29 13:10:36 +02001364 if (!(in->f_mode & FMODE_PREAD))
Jens Axboeb92ce552006-04-11 13:52:07 +02001365 return -EINVAL;
Jens Axboecbb7e572006-04-11 14:57:50 +02001366 if (copy_from_user(&offset, off_in, sizeof(loff_t)))
Jens Axboeb92ce552006-04-11 13:52:07 +02001367 return -EFAULT;
Jens Axboecbb7e572006-04-11 14:57:50 +02001368 off = &offset;
1369 } else
1370 off = &in->f_pos;
Ingo Molnar529565d2006-04-10 15:18:58 +02001371
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001372 ret = do_splice_to(in, off, opipe, len, flags);
Jens Axboea4514eb2006-04-19 15:57:05 +02001373
1374 if (off_in && copy_to_user(off_in, off, sizeof(loff_t)))
1375 ret = -EFAULT;
1376
1377 return ret;
Ingo Molnar529565d2006-04-10 15:18:58 +02001378 }
Jens Axboe5274f052006-03-30 15:15:30 +02001379
1380 return -EINVAL;
1381}
1382
Jens Axboe912d35f2006-04-26 10:59:21 +02001383/*
1384 * Map an iov into an array of pages and offset/length tupples. With the
1385 * partial_page structure, we can map several non-contiguous ranges into
1386 * our ones pages[] map instead of splitting that operation into pieces.
1387 * Could easily be exported as a generic helper for other users, in which
1388 * case one would probably want to add a 'max_nr_pages' parameter as well.
1389 */
1390static int get_iovec_page_array(const struct iovec __user *iov,
1391 unsigned int nr_vecs, struct page **pages,
Jens Axboe35f3d142010-05-20 10:43:18 +02001392 struct partial_page *partial, int aligned,
1393 unsigned int pipe_buffers)
Jens Axboe912d35f2006-04-26 10:59:21 +02001394{
1395 int buffers = 0, error = 0;
1396
Jens Axboe912d35f2006-04-26 10:59:21 +02001397 while (nr_vecs) {
1398 unsigned long off, npages;
Linus Torvalds75723952007-10-01 13:17:28 -07001399 struct iovec entry;
Jens Axboe912d35f2006-04-26 10:59:21 +02001400 void __user *base;
1401 size_t len;
1402 int i;
1403
Linus Torvalds75723952007-10-01 13:17:28 -07001404 error = -EFAULT;
Nick Pigginbc40d732008-07-25 19:45:26 -07001405 if (copy_from_user(&entry, iov, sizeof(entry)))
Jens Axboe912d35f2006-04-26 10:59:21 +02001406 break;
Linus Torvalds75723952007-10-01 13:17:28 -07001407
1408 base = entry.iov_base;
1409 len = entry.iov_len;
Jens Axboe912d35f2006-04-26 10:59:21 +02001410
1411 /*
1412 * Sanity check this iovec. 0 read succeeds.
1413 */
Linus Torvalds75723952007-10-01 13:17:28 -07001414 error = 0;
Jens Axboe912d35f2006-04-26 10:59:21 +02001415 if (unlikely(!len))
1416 break;
1417 error = -EFAULT;
Bastian Blank712a30e2008-02-10 16:47:57 +02001418 if (!access_ok(VERIFY_READ, base, len))
Jens Axboe912d35f2006-04-26 10:59:21 +02001419 break;
1420
1421 /*
1422 * Get this base offset and number of pages, then map
1423 * in the user pages.
1424 */
1425 off = (unsigned long) base & ~PAGE_MASK;
Jens Axboe7afa6fd2006-05-01 20:02:33 +02001426
1427 /*
1428 * If asked for alignment, the offset must be zero and the
1429 * length a multiple of the PAGE_SIZE.
1430 */
1431 error = -EINVAL;
1432 if (aligned && (off || len & ~PAGE_MASK))
1433 break;
1434
Jens Axboe912d35f2006-04-26 10:59:21 +02001435 npages = (off + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
Jens Axboe35f3d142010-05-20 10:43:18 +02001436 if (npages > pipe_buffers - buffers)
1437 npages = pipe_buffers - buffers;
Jens Axboe912d35f2006-04-26 10:59:21 +02001438
Nick Pigginbc40d732008-07-25 19:45:26 -07001439 error = get_user_pages_fast((unsigned long)base, npages,
1440 0, &pages[buffers]);
Jens Axboe912d35f2006-04-26 10:59:21 +02001441
1442 if (unlikely(error <= 0))
1443 break;
1444
1445 /*
1446 * Fill this contiguous range into the partial page map.
1447 */
1448 for (i = 0; i < error; i++) {
Jens Axboe75914892006-05-02 12:57:18 +02001449 const int plen = min_t(size_t, len, PAGE_SIZE - off);
Jens Axboe912d35f2006-04-26 10:59:21 +02001450
1451 partial[buffers].offset = off;
1452 partial[buffers].len = plen;
1453
1454 off = 0;
1455 len -= plen;
1456 buffers++;
1457 }
1458
1459 /*
1460 * We didn't complete this iov, stop here since it probably
1461 * means we have to move some of this into a pipe to
1462 * be able to continue.
1463 */
1464 if (len)
1465 break;
1466
1467 /*
1468 * Don't continue if we mapped fewer pages than we asked for,
1469 * or if we mapped the max number of pages that we have
1470 * room for.
1471 */
Jens Axboe35f3d142010-05-20 10:43:18 +02001472 if (error < npages || buffers == pipe_buffers)
Jens Axboe912d35f2006-04-26 10:59:21 +02001473 break;
1474
1475 nr_vecs--;
1476 iov++;
1477 }
1478
Jens Axboe912d35f2006-04-26 10:59:21 +02001479 if (buffers)
1480 return buffers;
1481
1482 return error;
1483}
1484
Jens Axboe6a14b902007-06-14 13:08:55 +02001485static int pipe_to_user(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
1486 struct splice_desc *sd)
1487{
1488 char *src;
1489 int ret;
1490
Jens Axboe6a14b902007-06-14 13:08:55 +02001491 /*
1492 * See if we can use the atomic maps, by prefaulting in the
1493 * pages and doing an atomic copy
1494 */
1495 if (!fault_in_pages_writeable(sd->u.userptr, sd->len)) {
1496 src = buf->ops->map(pipe, buf, 1);
1497 ret = __copy_to_user_inatomic(sd->u.userptr, src + buf->offset,
1498 sd->len);
1499 buf->ops->unmap(pipe, buf, src);
1500 if (!ret) {
1501 ret = sd->len;
1502 goto out;
1503 }
1504 }
1505
1506 /*
1507 * No dice, use slow non-atomic map and copy
1508 */
1509 src = buf->ops->map(pipe, buf, 0);
1510
1511 ret = sd->len;
1512 if (copy_to_user(sd->u.userptr, src + buf->offset, sd->len))
1513 ret = -EFAULT;
1514
Jens Axboe6866bef2007-10-16 10:01:29 +02001515 buf->ops->unmap(pipe, buf, src);
Jens Axboe6a14b902007-06-14 13:08:55 +02001516out:
1517 if (ret > 0)
1518 sd->u.userptr += ret;
Jens Axboe6a14b902007-06-14 13:08:55 +02001519 return ret;
1520}
1521
1522/*
1523 * For lack of a better implementation, implement vmsplice() to userspace
1524 * as a simple copy of the pipes pages to the user iov.
1525 */
1526static long vmsplice_to_user(struct file *file, const struct iovec __user *iov,
1527 unsigned long nr_segs, unsigned int flags)
1528{
1529 struct pipe_inode_info *pipe;
1530 struct splice_desc sd;
1531 ssize_t size;
1532 int error;
1533 long ret;
1534
Linus Torvalds71993e62010-11-28 13:56:09 -08001535 pipe = get_pipe_info(file);
Jens Axboe6a14b902007-06-14 13:08:55 +02001536 if (!pipe)
1537 return -EBADF;
1538
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001539 pipe_lock(pipe);
Jens Axboe6a14b902007-06-14 13:08:55 +02001540
1541 error = ret = 0;
1542 while (nr_segs) {
1543 void __user *base;
1544 size_t len;
1545
1546 /*
1547 * Get user address base and length for this iovec.
1548 */
1549 error = get_user(base, &iov->iov_base);
1550 if (unlikely(error))
1551 break;
1552 error = get_user(len, &iov->iov_len);
1553 if (unlikely(error))
1554 break;
1555
1556 /*
1557 * Sanity check this iovec. 0 read succeeds.
1558 */
1559 if (unlikely(!len))
1560 break;
1561 if (unlikely(!base)) {
1562 error = -EFAULT;
1563 break;
1564 }
1565
Jens Axboe88119302008-02-08 08:49:14 -08001566 if (unlikely(!access_ok(VERIFY_WRITE, base, len))) {
1567 error = -EFAULT;
1568 break;
1569 }
1570
Jens Axboe6a14b902007-06-14 13:08:55 +02001571 sd.len = 0;
1572 sd.total_len = len;
1573 sd.flags = flags;
1574 sd.u.userptr = base;
1575 sd.pos = 0;
1576
1577 size = __splice_from_pipe(pipe, &sd, pipe_to_user);
1578 if (size < 0) {
1579 if (!ret)
1580 ret = size;
1581
1582 break;
1583 }
1584
1585 ret += size;
1586
1587 if (size < len)
1588 break;
1589
1590 nr_segs--;
1591 iov++;
1592 }
1593
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001594 pipe_unlock(pipe);
Jens Axboe6a14b902007-06-14 13:08:55 +02001595
1596 if (!ret)
1597 ret = error;
1598
1599 return ret;
1600}
1601
Jens Axboe912d35f2006-04-26 10:59:21 +02001602/*
1603 * vmsplice splices a user address range into a pipe. It can be thought of
1604 * as splice-from-memory, where the regular splice is splice-from-file (or
1605 * to file). In both cases the output is a pipe, naturally.
Jens Axboe912d35f2006-04-26 10:59:21 +02001606 */
Jens Axboe6a14b902007-06-14 13:08:55 +02001607static long vmsplice_to_pipe(struct file *file, const struct iovec __user *iov,
1608 unsigned long nr_segs, unsigned int flags)
Jens Axboe912d35f2006-04-26 10:59:21 +02001609{
Jens Axboeddac0d32006-11-04 12:49:32 +01001610 struct pipe_inode_info *pipe;
Jens Axboe35f3d142010-05-20 10:43:18 +02001611 struct page *pages[PIPE_DEF_BUFFERS];
1612 struct partial_page partial[PIPE_DEF_BUFFERS];
Jens Axboe912d35f2006-04-26 10:59:21 +02001613 struct splice_pipe_desc spd = {
1614 .pages = pages,
1615 .partial = partial,
1616 .flags = flags,
1617 .ops = &user_page_pipe_buf_ops,
Jens Axboebbdfc2f2007-11-06 23:29:47 -08001618 .spd_release = spd_release_page,
Jens Axboe912d35f2006-04-26 10:59:21 +02001619 };
Jens Axboe35f3d142010-05-20 10:43:18 +02001620 long ret;
Jens Axboe912d35f2006-04-26 10:59:21 +02001621
Linus Torvalds71993e62010-11-28 13:56:09 -08001622 pipe = get_pipe_info(file);
Jens Axboeddac0d32006-11-04 12:49:32 +01001623 if (!pipe)
Jens Axboe912d35f2006-04-26 10:59:21 +02001624 return -EBADF;
Jens Axboe912d35f2006-04-26 10:59:21 +02001625
Jens Axboe35f3d142010-05-20 10:43:18 +02001626 if (splice_grow_spd(pipe, &spd))
1627 return -ENOMEM;
Jens Axboe912d35f2006-04-26 10:59:21 +02001628
Jens Axboe35f3d142010-05-20 10:43:18 +02001629 spd.nr_pages = get_iovec_page_array(iov, nr_segs, spd.pages,
1630 spd.partial, flags & SPLICE_F_GIFT,
1631 pipe->buffers);
1632 if (spd.nr_pages <= 0)
1633 ret = spd.nr_pages;
1634 else
1635 ret = splice_to_pipe(pipe, &spd);
1636
1637 splice_shrink_spd(pipe, &spd);
1638 return ret;
Jens Axboe912d35f2006-04-26 10:59:21 +02001639}
1640
Jens Axboe6a14b902007-06-14 13:08:55 +02001641/*
1642 * Note that vmsplice only really supports true splicing _from_ user memory
1643 * to a pipe, not the other way around. Splicing from user memory is a simple
1644 * operation that can be supported without any funky alignment restrictions
1645 * or nasty vm tricks. We simply map in the user memory and fill them into
1646 * a pipe. The reverse isn't quite as easy, though. There are two possible
1647 * solutions for that:
1648 *
1649 * - memcpy() the data internally, at which point we might as well just
1650 * do a regular read() on the buffer anyway.
1651 * - Lots of nasty vm tricks, that are neither fast nor flexible (it
1652 * has restriction limitations on both ends of the pipe).
1653 *
1654 * Currently we punt and implement it as a normal copy, see pipe_to_user().
1655 *
1656 */
Heiko Carstens836f92a2009-01-14 14:14:33 +01001657SYSCALL_DEFINE4(vmsplice, int, fd, const struct iovec __user *, iov,
1658 unsigned long, nr_segs, unsigned int, flags)
Jens Axboe912d35f2006-04-26 10:59:21 +02001659{
1660 struct file *file;
1661 long error;
1662 int fput;
1663
Jens Axboe6a14b902007-06-14 13:08:55 +02001664 if (unlikely(nr_segs > UIO_MAXIOV))
1665 return -EINVAL;
1666 else if (unlikely(!nr_segs))
1667 return 0;
1668
Jens Axboe912d35f2006-04-26 10:59:21 +02001669 error = -EBADF;
1670 file = fget_light(fd, &fput);
1671 if (file) {
1672 if (file->f_mode & FMODE_WRITE)
Jens Axboe6a14b902007-06-14 13:08:55 +02001673 error = vmsplice_to_pipe(file, iov, nr_segs, flags);
1674 else if (file->f_mode & FMODE_READ)
1675 error = vmsplice_to_user(file, iov, nr_segs, flags);
Jens Axboe912d35f2006-04-26 10:59:21 +02001676
1677 fput_light(file, fput);
1678 }
1679
1680 return error;
1681}
1682
Heiko Carstens836f92a2009-01-14 14:14:33 +01001683SYSCALL_DEFINE6(splice, int, fd_in, loff_t __user *, off_in,
1684 int, fd_out, loff_t __user *, off_out,
1685 size_t, len, unsigned int, flags)
Jens Axboe5274f052006-03-30 15:15:30 +02001686{
1687 long error;
1688 struct file *in, *out;
1689 int fput_in, fput_out;
1690
1691 if (unlikely(!len))
1692 return 0;
1693
1694 error = -EBADF;
Ingo Molnar529565d2006-04-10 15:18:58 +02001695 in = fget_light(fd_in, &fput_in);
Jens Axboe5274f052006-03-30 15:15:30 +02001696 if (in) {
1697 if (in->f_mode & FMODE_READ) {
Ingo Molnar529565d2006-04-10 15:18:58 +02001698 out = fget_light(fd_out, &fput_out);
Jens Axboe5274f052006-03-30 15:15:30 +02001699 if (out) {
1700 if (out->f_mode & FMODE_WRITE)
Ingo Molnar529565d2006-04-10 15:18:58 +02001701 error = do_splice(in, off_in,
1702 out, off_out,
1703 len, flags);
Jens Axboe5274f052006-03-30 15:15:30 +02001704 fput_light(out, fput_out);
1705 }
1706 }
1707
1708 fput_light(in, fput_in);
1709 }
1710
1711 return error;
1712}
Jens Axboe70524492006-04-11 15:51:17 +02001713
1714/*
Jens Axboeaadd06e2006-07-10 11:00:01 +02001715 * Make sure there's data to read. Wait for input if we can, otherwise
1716 * return an appropriate error.
1717 */
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001718static int ipipe_prep(struct pipe_inode_info *pipe, unsigned int flags)
Jens Axboeaadd06e2006-07-10 11:00:01 +02001719{
1720 int ret;
1721
1722 /*
1723 * Check ->nrbufs without the inode lock first. This function
1724 * is speculative anyways, so missing one is ok.
1725 */
1726 if (pipe->nrbufs)
1727 return 0;
1728
1729 ret = 0;
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001730 pipe_lock(pipe);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001731
1732 while (!pipe->nrbufs) {
1733 if (signal_pending(current)) {
1734 ret = -ERESTARTSYS;
1735 break;
1736 }
1737 if (!pipe->writers)
1738 break;
1739 if (!pipe->waiting_writers) {
1740 if (flags & SPLICE_F_NONBLOCK) {
1741 ret = -EAGAIN;
1742 break;
1743 }
1744 }
1745 pipe_wait(pipe);
1746 }
1747
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001748 pipe_unlock(pipe);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001749 return ret;
1750}
1751
1752/*
1753 * Make sure there's writeable room. Wait for room if we can, otherwise
1754 * return an appropriate error.
1755 */
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001756static int opipe_prep(struct pipe_inode_info *pipe, unsigned int flags)
Jens Axboeaadd06e2006-07-10 11:00:01 +02001757{
1758 int ret;
1759
1760 /*
1761 * Check ->nrbufs without the inode lock first. This function
1762 * is speculative anyways, so missing one is ok.
1763 */
Jens Axboe35f3d142010-05-20 10:43:18 +02001764 if (pipe->nrbufs < pipe->buffers)
Jens Axboeaadd06e2006-07-10 11:00:01 +02001765 return 0;
1766
1767 ret = 0;
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001768 pipe_lock(pipe);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001769
Jens Axboe35f3d142010-05-20 10:43:18 +02001770 while (pipe->nrbufs >= pipe->buffers) {
Jens Axboeaadd06e2006-07-10 11:00:01 +02001771 if (!pipe->readers) {
1772 send_sig(SIGPIPE, current, 0);
1773 ret = -EPIPE;
1774 break;
1775 }
1776 if (flags & SPLICE_F_NONBLOCK) {
1777 ret = -EAGAIN;
1778 break;
1779 }
1780 if (signal_pending(current)) {
1781 ret = -ERESTARTSYS;
1782 break;
1783 }
1784 pipe->waiting_writers++;
1785 pipe_wait(pipe);
1786 pipe->waiting_writers--;
1787 }
1788
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001789 pipe_unlock(pipe);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001790 return ret;
1791}
1792
1793/*
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001794 * Splice contents of ipipe to opipe.
1795 */
1796static int splice_pipe_to_pipe(struct pipe_inode_info *ipipe,
1797 struct pipe_inode_info *opipe,
1798 size_t len, unsigned int flags)
1799{
1800 struct pipe_buffer *ibuf, *obuf;
1801 int ret = 0, nbuf;
1802 bool input_wakeup = false;
1803
1804
1805retry:
1806 ret = ipipe_prep(ipipe, flags);
1807 if (ret)
1808 return ret;
1809
1810 ret = opipe_prep(opipe, flags);
1811 if (ret)
1812 return ret;
1813
1814 /*
1815 * Potential ABBA deadlock, work around it by ordering lock
1816 * grabbing by pipe info address. Otherwise two different processes
1817 * could deadlock (one doing tee from A -> B, the other from B -> A).
1818 */
1819 pipe_double_lock(ipipe, opipe);
1820
1821 do {
1822 if (!opipe->readers) {
1823 send_sig(SIGPIPE, current, 0);
1824 if (!ret)
1825 ret = -EPIPE;
1826 break;
1827 }
1828
1829 if (!ipipe->nrbufs && !ipipe->writers)
1830 break;
1831
1832 /*
1833 * Cannot make any progress, because either the input
1834 * pipe is empty or the output pipe is full.
1835 */
Jens Axboe35f3d142010-05-20 10:43:18 +02001836 if (!ipipe->nrbufs || opipe->nrbufs >= opipe->buffers) {
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001837 /* Already processed some buffers, break */
1838 if (ret)
1839 break;
1840
1841 if (flags & SPLICE_F_NONBLOCK) {
1842 ret = -EAGAIN;
1843 break;
1844 }
1845
1846 /*
1847 * We raced with another reader/writer and haven't
1848 * managed to process any buffers. A zero return
1849 * value means EOF, so retry instead.
1850 */
1851 pipe_unlock(ipipe);
1852 pipe_unlock(opipe);
1853 goto retry;
1854 }
1855
1856 ibuf = ipipe->bufs + ipipe->curbuf;
Jens Axboe35f3d142010-05-20 10:43:18 +02001857 nbuf = (opipe->curbuf + opipe->nrbufs) & (opipe->buffers - 1);
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001858 obuf = opipe->bufs + nbuf;
1859
1860 if (len >= ibuf->len) {
1861 /*
1862 * Simply move the whole buffer from ipipe to opipe
1863 */
1864 *obuf = *ibuf;
1865 ibuf->ops = NULL;
1866 opipe->nrbufs++;
Jens Axboe35f3d142010-05-20 10:43:18 +02001867 ipipe->curbuf = (ipipe->curbuf + 1) & (ipipe->buffers - 1);
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001868 ipipe->nrbufs--;
1869 input_wakeup = true;
1870 } else {
1871 /*
1872 * Get a reference to this pipe buffer,
1873 * so we can copy the contents over.
1874 */
1875 ibuf->ops->get(ipipe, ibuf);
1876 *obuf = *ibuf;
1877
1878 /*
1879 * Don't inherit the gift flag, we need to
1880 * prevent multiple steals of this page.
1881 */
1882 obuf->flags &= ~PIPE_BUF_FLAG_GIFT;
1883
1884 obuf->len = len;
1885 opipe->nrbufs++;
1886 ibuf->offset += obuf->len;
1887 ibuf->len -= obuf->len;
1888 }
1889 ret += obuf->len;
1890 len -= obuf->len;
1891 } while (len);
1892
1893 pipe_unlock(ipipe);
1894 pipe_unlock(opipe);
1895
1896 /*
1897 * If we put data in the output pipe, wakeup any potential readers.
1898 */
Namhyung Kim825cdcb2011-05-23 19:58:53 +02001899 if (ret > 0)
1900 wakeup_pipe_readers(opipe);
1901
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001902 if (input_wakeup)
1903 wakeup_pipe_writers(ipipe);
1904
1905 return ret;
1906}
1907
1908/*
Jens Axboe70524492006-04-11 15:51:17 +02001909 * Link contents of ipipe to opipe.
1910 */
1911static int link_pipe(struct pipe_inode_info *ipipe,
1912 struct pipe_inode_info *opipe,
1913 size_t len, unsigned int flags)
1914{
1915 struct pipe_buffer *ibuf, *obuf;
Jens Axboeaadd06e2006-07-10 11:00:01 +02001916 int ret = 0, i = 0, nbuf;
Jens Axboe70524492006-04-11 15:51:17 +02001917
1918 /*
1919 * Potential ABBA deadlock, work around it by ordering lock
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001920 * grabbing by pipe info address. Otherwise two different processes
Jens Axboe70524492006-04-11 15:51:17 +02001921 * could deadlock (one doing tee from A -> B, the other from B -> A).
1922 */
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001923 pipe_double_lock(ipipe, opipe);
Jens Axboe70524492006-04-11 15:51:17 +02001924
Jens Axboeaadd06e2006-07-10 11:00:01 +02001925 do {
Jens Axboe70524492006-04-11 15:51:17 +02001926 if (!opipe->readers) {
1927 send_sig(SIGPIPE, current, 0);
1928 if (!ret)
1929 ret = -EPIPE;
1930 break;
1931 }
Jens Axboe70524492006-04-11 15:51:17 +02001932
1933 /*
Jens Axboeaadd06e2006-07-10 11:00:01 +02001934 * If we have iterated all input buffers or ran out of
1935 * output room, break.
Jens Axboe70524492006-04-11 15:51:17 +02001936 */
Jens Axboe35f3d142010-05-20 10:43:18 +02001937 if (i >= ipipe->nrbufs || opipe->nrbufs >= opipe->buffers)
Jens Axboe70524492006-04-11 15:51:17 +02001938 break;
Jens Axboeaadd06e2006-07-10 11:00:01 +02001939
Jens Axboe35f3d142010-05-20 10:43:18 +02001940 ibuf = ipipe->bufs + ((ipipe->curbuf + i) & (ipipe->buffers-1));
1941 nbuf = (opipe->curbuf + opipe->nrbufs) & (opipe->buffers - 1);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001942
Jens Axboe2a272502006-04-19 15:56:40 +02001943 /*
Jens Axboeaadd06e2006-07-10 11:00:01 +02001944 * Get a reference to this pipe buffer,
1945 * so we can copy the contents over.
Jens Axboe2a272502006-04-19 15:56:40 +02001946 */
Jens Axboeaadd06e2006-07-10 11:00:01 +02001947 ibuf->ops->get(ipipe, ibuf);
Jens Axboe70524492006-04-11 15:51:17 +02001948
Jens Axboeaadd06e2006-07-10 11:00:01 +02001949 obuf = opipe->bufs + nbuf;
1950 *obuf = *ibuf;
Jens Axboe70524492006-04-11 15:51:17 +02001951
Jens Axboeaadd06e2006-07-10 11:00:01 +02001952 /*
1953 * Don't inherit the gift flag, we need to
1954 * prevent multiple steals of this page.
1955 */
1956 obuf->flags &= ~PIPE_BUF_FLAG_GIFT;
1957
1958 if (obuf->len > len)
1959 obuf->len = len;
1960
1961 opipe->nrbufs++;
1962 ret += obuf->len;
1963 len -= obuf->len;
1964 i++;
1965 } while (len);
Jens Axboe70524492006-04-11 15:51:17 +02001966
Jens Axboe02cf01a2008-02-20 10:34:51 +01001967 /*
1968 * return EAGAIN if we have the potential of some data in the
1969 * future, otherwise just return 0
1970 */
1971 if (!ret && ipipe->waiting_writers && (flags & SPLICE_F_NONBLOCK))
1972 ret = -EAGAIN;
1973
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001974 pipe_unlock(ipipe);
1975 pipe_unlock(opipe);
Jens Axboe70524492006-04-11 15:51:17 +02001976
Jens Axboeaadd06e2006-07-10 11:00:01 +02001977 /*
1978 * If we put data in the output pipe, wakeup any potential readers.
1979 */
Namhyung Kim825cdcb2011-05-23 19:58:53 +02001980 if (ret > 0)
1981 wakeup_pipe_readers(opipe);
Jens Axboe70524492006-04-11 15:51:17 +02001982
1983 return ret;
1984}
1985
1986/*
1987 * This is a tee(1) implementation that works on pipes. It doesn't copy
1988 * any data, it simply references the 'in' pages on the 'out' pipe.
1989 * The 'flags' used are the SPLICE_F_* variants, currently the only
1990 * applicable one is SPLICE_F_NONBLOCK.
1991 */
1992static long do_tee(struct file *in, struct file *out, size_t len,
1993 unsigned int flags)
1994{
Linus Torvalds71993e62010-11-28 13:56:09 -08001995 struct pipe_inode_info *ipipe = get_pipe_info(in);
1996 struct pipe_inode_info *opipe = get_pipe_info(out);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001997 int ret = -EINVAL;
Jens Axboe70524492006-04-11 15:51:17 +02001998
1999 /*
Jens Axboeaadd06e2006-07-10 11:00:01 +02002000 * Duplicate the contents of ipipe to opipe without actually
2001 * copying the data.
Jens Axboe70524492006-04-11 15:51:17 +02002002 */
Jens Axboeaadd06e2006-07-10 11:00:01 +02002003 if (ipipe && opipe && ipipe != opipe) {
2004 /*
2005 * Keep going, unless we encounter an error. The ipipe/opipe
2006 * ordering doesn't really matter.
2007 */
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02002008 ret = ipipe_prep(ipipe, flags);
Jens Axboeaadd06e2006-07-10 11:00:01 +02002009 if (!ret) {
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02002010 ret = opipe_prep(opipe, flags);
Jens Axboe02cf01a2008-02-20 10:34:51 +01002011 if (!ret)
Jens Axboeaadd06e2006-07-10 11:00:01 +02002012 ret = link_pipe(ipipe, opipe, len, flags);
Jens Axboeaadd06e2006-07-10 11:00:01 +02002013 }
2014 }
Jens Axboe70524492006-04-11 15:51:17 +02002015
Jens Axboeaadd06e2006-07-10 11:00:01 +02002016 return ret;
Jens Axboe70524492006-04-11 15:51:17 +02002017}
2018
Heiko Carstens836f92a2009-01-14 14:14:33 +01002019SYSCALL_DEFINE4(tee, int, fdin, int, fdout, size_t, len, unsigned int, flags)
Jens Axboe70524492006-04-11 15:51:17 +02002020{
2021 struct file *in;
2022 int error, fput_in;
2023
2024 if (unlikely(!len))
2025 return 0;
2026
2027 error = -EBADF;
2028 in = fget_light(fdin, &fput_in);
2029 if (in) {
2030 if (in->f_mode & FMODE_READ) {
2031 int fput_out;
2032 struct file *out = fget_light(fdout, &fput_out);
2033
2034 if (out) {
2035 if (out->f_mode & FMODE_WRITE)
2036 error = do_tee(in, out, len, flags);
2037 fput_light(out, fput_out);
2038 }
2039 }
2040 fput_light(in, fput_in);
2041 }
2042
2043 return error;
2044}