blob: f8476841eb04e08edc2ad069e97cddc3da242e7f [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 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
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 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
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 */
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
Eric Dumazet35f9c092012-04-05 03:05:35 +0000694 more = (sd->flags & SPLICE_F_MORE) ? MSG_MORE : 0;
695 if (sd->len < sd->total_len)
696 more |= MSG_SENDPAGE_NOTLAST;
Michał Mirosława8adbe32010-12-17 08:56:44 +0100697 return file->f_op->sendpage(file, buf->page, buf->offset,
698 sd->len, &pos, more);
Jens Axboe5274f052006-03-30 15:15:30 +0200699}
700
701/*
702 * This is a little more tricky than the file -> pipe splicing. There are
703 * basically three cases:
704 *
705 * - Destination page already exists in the address space and there
706 * are users of it. For that case we have no other option that
707 * copying the data. Tough luck.
708 * - Destination page already exists in the address space, but there
709 * are no users of it. Make sure it's uptodate, then drop it. Fall
710 * through to last case.
711 * - Destination page does not exist, we can add the pipe page to
712 * the page cache and avoid the copy.
713 *
Jens Axboe83f91352006-04-02 23:05:09 +0200714 * If asked to move pages to the output file (SPLICE_F_MOVE is set in
715 * sd->flags), we attempt to migrate pages from the pipe to the output
716 * file address space page cache. This is possible if no one else has
717 * the pipe page referenced outside of the pipe and page cache. If
718 * SPLICE_F_MOVE isn't set, or we cannot move the page, we simply create
719 * a new page in the output file page cache and fill/dirty that.
Jens Axboe5274f052006-03-30 15:15:30 +0200720 */
Miklos Szeredi328eaab2009-04-14 19:48:39 +0200721int pipe_to_file(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
722 struct splice_desc *sd)
Jens Axboe5274f052006-03-30 15:15:30 +0200723{
Jens Axboe6a14b902007-06-14 13:08:55 +0200724 struct file *file = sd->u.file;
Jens Axboe5274f052006-03-30 15:15:30 +0200725 struct address_space *mapping = file->f_mapping;
Jens Axboe016b6612006-04-25 15:42:00 +0200726 unsigned int offset, this_len;
Jens Axboe5274f052006-03-30 15:15:30 +0200727 struct page *page;
Nick Pigginafddba42007-10-16 01:25:01 -0700728 void *fsdata;
Jens Axboe3e7ee3e2006-04-02 23:11:04 +0200729 int ret;
Jens Axboe5274f052006-03-30 15:15:30 +0200730
Jens Axboe5274f052006-03-30 15:15:30 +0200731 offset = sd->pos & ~PAGE_CACHE_MASK;
732
Jens Axboe016b6612006-04-25 15:42:00 +0200733 this_len = sd->len;
734 if (this_len + offset > PAGE_CACHE_SIZE)
735 this_len = PAGE_CACHE_SIZE - offset;
736
Nick Pigginafddba42007-10-16 01:25:01 -0700737 ret = pagecache_write_begin(file, mapping, sd->pos, this_len,
738 AOP_FLAG_UNINTERRUPTIBLE, &page, &fsdata);
739 if (unlikely(ret))
740 goto out;
Jens Axboe5274f052006-03-30 15:15:30 +0200741
Jens Axboe0568b402006-05-01 19:50:48 +0200742 if (buf->page != page) {
Jens Axboe76ad4d12006-05-03 10:41:33 +0200743 char *src = buf->ops->map(pipe, buf, 1);
Cong Wange8e3c3d2011-11-25 23:14:27 +0800744 char *dst = kmap_atomic(page);
Jens Axboe5abc97a2006-03-30 15:16:46 +0200745
Jens Axboe016b6612006-04-25 15:42:00 +0200746 memcpy(dst + offset, src + buf->offset, this_len);
Jens Axboe5abc97a2006-03-30 15:16:46 +0200747 flush_dcache_page(page);
Cong Wange8e3c3d2011-11-25 23:14:27 +0800748 kunmap_atomic(dst);
Jens Axboe76ad4d12006-05-03 10:41:33 +0200749 buf->ops->unmap(pipe, buf, src);
Jens Axboe5abc97a2006-03-30 15:16:46 +0200750 }
Nick Pigginafddba42007-10-16 01:25:01 -0700751 ret = pagecache_write_end(file, mapping, sd->pos, this_len, this_len,
752 page, fsdata);
Jens Axboe5274f052006-03-30 15:15:30 +0200753out:
Jens Axboe5274f052006-03-30 15:15:30 +0200754 return ret;
755}
Miklos Szeredi328eaab2009-04-14 19:48:39 +0200756EXPORT_SYMBOL(pipe_to_file);
Jens Axboe5274f052006-03-30 15:15:30 +0200757
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200758static void wakeup_pipe_writers(struct pipe_inode_info *pipe)
759{
760 smp_mb();
761 if (waitqueue_active(&pipe->wait))
762 wake_up_interruptible(&pipe->wait);
763 kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
764}
765
766/**
767 * splice_from_pipe_feed - feed available data from a pipe to a file
768 * @pipe: pipe to splice from
769 * @sd: information to @actor
770 * @actor: handler that splices the data
771 *
772 * Description:
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200773 * This function loops over the pipe and calls @actor to do the
774 * actual moving of a single struct pipe_buffer to the desired
775 * destination. It returns when there's no more buffers left in
776 * the pipe or if the requested number of bytes (@sd->total_len)
777 * have been copied. It returns a positive number (one) if the
778 * pipe needs to be filled with more data, zero if the required
779 * number of bytes have been copied and -errno on error.
780 *
781 * This, together with splice_from_pipe_{begin,end,next}, may be
782 * used to implement the functionality of __splice_from_pipe() when
783 * locking is required around copying the pipe buffers to the
784 * destination.
785 */
786int splice_from_pipe_feed(struct pipe_inode_info *pipe, struct splice_desc *sd,
787 splice_actor *actor)
788{
789 int ret;
790
791 while (pipe->nrbufs) {
792 struct pipe_buffer *buf = pipe->bufs + pipe->curbuf;
793 const struct pipe_buf_operations *ops = buf->ops;
794
795 sd->len = buf->len;
796 if (sd->len > sd->total_len)
797 sd->len = sd->total_len;
798
Michał Mirosława8adbe32010-12-17 08:56:44 +0100799 ret = buf->ops->confirm(pipe, buf);
800 if (unlikely(ret)) {
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200801 if (ret == -ENODATA)
802 ret = 0;
803 return ret;
804 }
Michał Mirosława8adbe32010-12-17 08:56:44 +0100805
806 ret = actor(pipe, buf, sd);
807 if (ret <= 0)
808 return ret;
809
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200810 buf->offset += ret;
811 buf->len -= ret;
812
813 sd->num_spliced += ret;
814 sd->len -= ret;
815 sd->pos += ret;
816 sd->total_len -= ret;
817
818 if (!buf->len) {
819 buf->ops = NULL;
820 ops->release(pipe, buf);
Jens Axboe35f3d142010-05-20 10:43:18 +0200821 pipe->curbuf = (pipe->curbuf + 1) & (pipe->buffers - 1);
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200822 pipe->nrbufs--;
823 if (pipe->inode)
824 sd->need_wakeup = true;
825 }
826
827 if (!sd->total_len)
828 return 0;
829 }
830
831 return 1;
832}
833EXPORT_SYMBOL(splice_from_pipe_feed);
834
835/**
836 * splice_from_pipe_next - wait for some data to splice from
837 * @pipe: pipe to splice from
838 * @sd: information about the splice operation
839 *
840 * Description:
841 * This function will wait for some data and return a positive
842 * value (one) if pipe buffers are available. It will return zero
843 * or -errno if no more data needs to be spliced.
844 */
845int splice_from_pipe_next(struct pipe_inode_info *pipe, struct splice_desc *sd)
846{
847 while (!pipe->nrbufs) {
848 if (!pipe->writers)
849 return 0;
850
851 if (!pipe->waiting_writers && sd->num_spliced)
852 return 0;
853
854 if (sd->flags & SPLICE_F_NONBLOCK)
855 return -EAGAIN;
856
857 if (signal_pending(current))
858 return -ERESTARTSYS;
859
860 if (sd->need_wakeup) {
861 wakeup_pipe_writers(pipe);
862 sd->need_wakeup = false;
863 }
864
865 pipe_wait(pipe);
866 }
867
868 return 1;
869}
870EXPORT_SYMBOL(splice_from_pipe_next);
871
872/**
873 * splice_from_pipe_begin - start splicing from pipe
Randy Dunlapb80901b2009-04-16 19:09:55 -0700874 * @sd: information about the splice operation
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200875 *
876 * Description:
877 * This function should be called before a loop containing
878 * splice_from_pipe_next() and splice_from_pipe_feed() to
879 * initialize the necessary fields of @sd.
880 */
881void splice_from_pipe_begin(struct splice_desc *sd)
882{
883 sd->num_spliced = 0;
884 sd->need_wakeup = false;
885}
886EXPORT_SYMBOL(splice_from_pipe_begin);
887
888/**
889 * splice_from_pipe_end - finish splicing from pipe
890 * @pipe: pipe to splice from
891 * @sd: information about the splice operation
892 *
893 * Description:
894 * This function will wake up pipe writers if necessary. It should
895 * be called after a loop containing splice_from_pipe_next() and
896 * splice_from_pipe_feed().
897 */
898void splice_from_pipe_end(struct pipe_inode_info *pipe, struct splice_desc *sd)
899{
900 if (sd->need_wakeup)
901 wakeup_pipe_writers(pipe);
902}
903EXPORT_SYMBOL(splice_from_pipe_end);
904
Jens Axboe932cc6d2007-06-21 13:10:21 +0200905/**
906 * __splice_from_pipe - splice data from a pipe to given actor
907 * @pipe: pipe to splice from
908 * @sd: information to @actor
909 * @actor: handler that splices the data
910 *
911 * Description:
912 * This function does little more than loop over the pipe and call
913 * @actor to do the actual moving of a single struct pipe_buffer to
914 * the desired destination. See pipe_to_file, pipe_to_sendpage, or
915 * pipe_to_user.
916 *
Jens Axboe83f91352006-04-02 23:05:09 +0200917 */
Jens Axboec66ab6f2007-06-12 21:17:17 +0200918ssize_t __splice_from_pipe(struct pipe_inode_info *pipe, struct splice_desc *sd,
919 splice_actor *actor)
Jens Axboe5274f052006-03-30 15:15:30 +0200920{
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200921 int ret;
Jens Axboe5274f052006-03-30 15:15:30 +0200922
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200923 splice_from_pipe_begin(sd);
924 do {
925 ret = splice_from_pipe_next(pipe, sd);
926 if (ret > 0)
927 ret = splice_from_pipe_feed(pipe, sd, actor);
928 } while (ret > 0);
929 splice_from_pipe_end(pipe, sd);
Jens Axboe5274f052006-03-30 15:15:30 +0200930
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200931 return sd->num_spliced ? sd->num_spliced : ret;
Jens Axboe5274f052006-03-30 15:15:30 +0200932}
Mark Fasheh40bee442007-03-21 13:11:02 +0100933EXPORT_SYMBOL(__splice_from_pipe);
Jens Axboe5274f052006-03-30 15:15:30 +0200934
Jens Axboe932cc6d2007-06-21 13:10:21 +0200935/**
936 * splice_from_pipe - splice data from a pipe to a file
937 * @pipe: pipe to splice from
938 * @out: file to splice to
939 * @ppos: position in @out
940 * @len: how many bytes to splice
941 * @flags: splice modifier flags
942 * @actor: handler that splices the data
943 *
944 * Description:
Miklos Szeredi29339702009-04-14 19:48:37 +0200945 * See __splice_from_pipe. This function locks the pipe inode,
Jens Axboe932cc6d2007-06-21 13:10:21 +0200946 * otherwise it's identical to __splice_from_pipe().
947 *
948 */
Mark Fasheh6da61802006-10-17 18:43:07 +0200949ssize_t splice_from_pipe(struct pipe_inode_info *pipe, struct file *out,
950 loff_t *ppos, size_t len, unsigned int flags,
951 splice_actor *actor)
952{
953 ssize_t ret;
Jens Axboec66ab6f2007-06-12 21:17:17 +0200954 struct splice_desc sd = {
955 .total_len = len,
956 .flags = flags,
957 .pos = *ppos,
Jens Axboe6a14b902007-06-14 13:08:55 +0200958 .u.file = out,
Jens Axboec66ab6f2007-06-12 21:17:17 +0200959 };
Mark Fasheh6da61802006-10-17 18:43:07 +0200960
Miklos Szeredi61e0d472009-04-14 19:48:41 +0200961 pipe_lock(pipe);
Jens Axboec66ab6f2007-06-12 21:17:17 +0200962 ret = __splice_from_pipe(pipe, &sd, actor);
Miklos Szeredi61e0d472009-04-14 19:48:41 +0200963 pipe_unlock(pipe);
Mark Fasheh6da61802006-10-17 18:43:07 +0200964
965 return ret;
966}
967
968/**
Jens Axboe83f91352006-04-02 23:05:09 +0200969 * generic_file_splice_write - splice data from a pipe to a file
Ingo Molnar3a326a22006-04-10 15:18:35 +0200970 * @pipe: pipe info
Jens Axboe83f91352006-04-02 23:05:09 +0200971 * @out: file to write to
Jens Axboe932cc6d2007-06-21 13:10:21 +0200972 * @ppos: position in @out
Jens Axboe83f91352006-04-02 23:05:09 +0200973 * @len: number of bytes to splice
974 * @flags: splice modifier flags
975 *
Jens Axboe932cc6d2007-06-21 13:10:21 +0200976 * Description:
977 * Will either move or copy pages (determined by @flags options) from
978 * the given pipe inode to the given file.
Jens Axboe83f91352006-04-02 23:05:09 +0200979 *
980 */
Ingo Molnar3a326a22006-04-10 15:18:35 +0200981ssize_t
982generic_file_splice_write(struct pipe_inode_info *pipe, struct file *out,
Jens Axboecbb7e572006-04-11 14:57:50 +0200983 loff_t *ppos, size_t len, unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +0200984{
Jens Axboe4f6f0bd2006-04-02 23:04:46 +0200985 struct address_space *mapping = out->f_mapping;
Jens Axboe8c34e2d2006-10-17 19:43:22 +0200986 struct inode *inode = mapping->host;
Miklos Szeredi7f3d4ee2008-05-07 09:22:39 +0200987 struct splice_desc sd = {
988 .total_len = len,
989 .flags = flags,
990 .pos = *ppos,
991 .u.file = out,
992 };
Ingo Molnar3a326a22006-04-10 15:18:35 +0200993 ssize_t ret;
Jens Axboe8c34e2d2006-10-17 19:43:22 +0200994
Miklos Szeredi61e0d472009-04-14 19:48:41 +0200995 pipe_lock(pipe);
Miklos Szeredieb443e52009-04-14 19:48:38 +0200996
997 splice_from_pipe_begin(&sd);
998 do {
999 ret = splice_from_pipe_next(pipe, &sd);
1000 if (ret <= 0)
1001 break;
1002
1003 mutex_lock_nested(&inode->i_mutex, I_MUTEX_CHILD);
1004 ret = file_remove_suid(out);
Miklos Szeredi723590e2009-08-15 08:43:22 +02001005 if (!ret) {
1006 file_update_time(out);
Miklos Szeredieb443e52009-04-14 19:48:38 +02001007 ret = splice_from_pipe_feed(pipe, &sd, pipe_to_file);
Miklos Szeredi723590e2009-08-15 08:43:22 +02001008 }
Miklos Szeredieb443e52009-04-14 19:48:38 +02001009 mutex_unlock(&inode->i_mutex);
1010 } while (ret > 0);
1011 splice_from_pipe_end(pipe, &sd);
1012
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001013 pipe_unlock(pipe);
Miklos Szeredieb443e52009-04-14 19:48:38 +02001014
1015 if (sd.num_spliced)
1016 ret = sd.num_spliced;
1017
Jens Axboea4514eb2006-04-19 15:57:05 +02001018 if (ret > 0) {
Jens Axboe17ee4f42007-06-15 13:10:37 +02001019 unsigned long nr_pages;
Jan Kara148f9482009-08-17 19:52:36 +02001020 int err;
Jens Axboe17ee4f42007-06-15 13:10:37 +02001021
Jens Axboe17ee4f42007-06-15 13:10:37 +02001022 nr_pages = (ret + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
Jens Axboe4f6f0bd2006-04-02 23:04:46 +02001023
Jan Kara148f9482009-08-17 19:52:36 +02001024 err = generic_write_sync(out, *ppos, ret);
1025 if (err)
1026 ret = err;
1027 else
1028 *ppos += ret;
Jens Axboe17ee4f42007-06-15 13:10:37 +02001029 balance_dirty_pages_ratelimited_nr(mapping, nr_pages);
Jens Axboe4f6f0bd2006-04-02 23:04:46 +02001030 }
1031
1032 return ret;
Jens Axboe5274f052006-03-30 15:15:30 +02001033}
1034
Jens Axboe059a8f32006-04-02 23:06:05 +02001035EXPORT_SYMBOL(generic_file_splice_write);
1036
Miklos Szeredib2858d72009-05-19 11:37:46 +02001037static int write_pipe_buf(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
1038 struct splice_desc *sd)
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +02001039{
Miklos Szeredib2858d72009-05-19 11:37:46 +02001040 int ret;
1041 void *data;
1042
Miklos Szeredib2858d72009-05-19 11:37:46 +02001043 data = buf->ops->map(pipe, buf, 0);
1044 ret = kernel_write(sd->u.file, data + buf->offset, sd->len, sd->pos);
1045 buf->ops->unmap(pipe, buf, data);
1046
1047 return ret;
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +02001048}
1049
1050static ssize_t default_file_splice_write(struct pipe_inode_info *pipe,
1051 struct file *out, loff_t *ppos,
1052 size_t len, unsigned int flags)
1053{
Miklos Szeredib2858d72009-05-19 11:37:46 +02001054 ssize_t ret;
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +02001055
Miklos Szeredib2858d72009-05-19 11:37:46 +02001056 ret = splice_from_pipe(pipe, out, ppos, len, flags, write_pipe_buf);
1057 if (ret > 0)
1058 *ppos += ret;
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +02001059
Miklos Szeredib2858d72009-05-19 11:37:46 +02001060 return ret;
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +02001061}
1062
Jens Axboe83f91352006-04-02 23:05:09 +02001063/**
1064 * generic_splice_sendpage - splice data from a pipe to a socket
Jens Axboe932cc6d2007-06-21 13:10:21 +02001065 * @pipe: pipe to splice from
Jens Axboe83f91352006-04-02 23:05:09 +02001066 * @out: socket to write to
Jens Axboe932cc6d2007-06-21 13:10:21 +02001067 * @ppos: position in @out
Jens Axboe83f91352006-04-02 23:05:09 +02001068 * @len: number of bytes to splice
1069 * @flags: splice modifier flags
1070 *
Jens Axboe932cc6d2007-06-21 13:10:21 +02001071 * Description:
1072 * Will send @len bytes from the pipe to a network socket. No data copying
1073 * is involved.
Jens Axboe83f91352006-04-02 23:05:09 +02001074 *
1075 */
Ingo Molnar3a326a22006-04-10 15:18:35 +02001076ssize_t generic_splice_sendpage(struct pipe_inode_info *pipe, struct file *out,
Jens Axboecbb7e572006-04-11 14:57:50 +02001077 loff_t *ppos, size_t len, unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +02001078{
Jens Axboe00522fb2006-04-26 14:39:29 +02001079 return splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_sendpage);
Jens Axboe5274f052006-03-30 15:15:30 +02001080}
1081
Jens Axboe059a8f32006-04-02 23:06:05 +02001082EXPORT_SYMBOL(generic_splice_sendpage);
Jeff Garzika0f06782006-03-30 23:06:13 -05001083
Jens Axboe83f91352006-04-02 23:05:09 +02001084/*
1085 * Attempt to initiate a splice from pipe to file.
1086 */
Ingo Molnar3a326a22006-04-10 15:18:35 +02001087static long do_splice_from(struct pipe_inode_info *pipe, struct file *out,
Jens Axboecbb7e572006-04-11 14:57:50 +02001088 loff_t *ppos, size_t len, unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +02001089{
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +02001090 ssize_t (*splice_write)(struct pipe_inode_info *, struct file *,
1091 loff_t *, size_t, unsigned int);
Jens Axboe5274f052006-03-30 15:15:30 +02001092 int ret;
1093
Jens Axboe49570e92006-04-11 13:56:09 +02001094 if (unlikely(!(out->f_mode & FMODE_WRITE)))
Jens Axboe5274f052006-03-30 15:15:30 +02001095 return -EBADF;
1096
Linus Torvaldsefc968d2008-10-09 14:04:54 -07001097 if (unlikely(out->f_flags & O_APPEND))
1098 return -EINVAL;
1099
Jens Axboecbb7e572006-04-11 14:57:50 +02001100 ret = rw_verify_area(WRITE, out, ppos, len);
Jens Axboe5274f052006-03-30 15:15:30 +02001101 if (unlikely(ret < 0))
1102 return ret;
1103
Changli Gaocc56f7d2009-11-04 09:09:52 +01001104 if (out->f_op && out->f_op->splice_write)
1105 splice_write = out->f_op->splice_write;
1106 else
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +02001107 splice_write = default_file_splice_write;
1108
1109 return splice_write(pipe, out, ppos, len, flags);
Jens Axboe5274f052006-03-30 15:15:30 +02001110}
1111
Jens Axboe83f91352006-04-02 23:05:09 +02001112/*
1113 * Attempt to initiate a splice from a file to a pipe.
1114 */
Jens Axboecbb7e572006-04-11 14:57:50 +02001115static long do_splice_to(struct file *in, loff_t *ppos,
1116 struct pipe_inode_info *pipe, size_t len,
1117 unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +02001118{
Miklos Szeredi68181732009-05-07 15:37:36 +02001119 ssize_t (*splice_read)(struct file *, loff_t *,
1120 struct pipe_inode_info *, size_t, unsigned int);
Jens Axboe5274f052006-03-30 15:15:30 +02001121 int ret;
1122
Jens Axboe49570e92006-04-11 13:56:09 +02001123 if (unlikely(!(in->f_mode & FMODE_READ)))
Jens Axboe5274f052006-03-30 15:15:30 +02001124 return -EBADF;
1125
Jens Axboecbb7e572006-04-11 14:57:50 +02001126 ret = rw_verify_area(READ, in, ppos, len);
Jens Axboe5274f052006-03-30 15:15:30 +02001127 if (unlikely(ret < 0))
1128 return ret;
1129
Changli Gaocc56f7d2009-11-04 09:09:52 +01001130 if (in->f_op && in->f_op->splice_read)
1131 splice_read = in->f_op->splice_read;
1132 else
Miklos Szeredi68181732009-05-07 15:37:36 +02001133 splice_read = default_file_splice_read;
1134
1135 return splice_read(in, ppos, pipe, len, flags);
Jens Axboe5274f052006-03-30 15:15:30 +02001136}
1137
Jens Axboe932cc6d2007-06-21 13:10:21 +02001138/**
1139 * splice_direct_to_actor - splices data directly between two non-pipes
1140 * @in: file to splice from
1141 * @sd: actor information on where to splice to
1142 * @actor: handles the data splicing
1143 *
1144 * Description:
1145 * This is a special case helper to splice directly between two
1146 * points, without requiring an explicit pipe. Internally an allocated
Randy Dunlap79685b82007-07-27 08:08:51 +02001147 * pipe is cached in the process, and reused during the lifetime of
Jens Axboe932cc6d2007-06-21 13:10:21 +02001148 * that process.
1149 *
Jens Axboec66ab6f2007-06-12 21:17:17 +02001150 */
1151ssize_t splice_direct_to_actor(struct file *in, struct splice_desc *sd,
1152 splice_direct_actor *actor)
Jens Axboeb92ce552006-04-11 13:52:07 +02001153{
1154 struct pipe_inode_info *pipe;
1155 long ret, bytes;
1156 umode_t i_mode;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001157 size_t len;
1158 int i, flags;
Jens Axboeb92ce552006-04-11 13:52:07 +02001159
1160 /*
1161 * We require the input being a regular file, as we don't want to
1162 * randomly drop data for eg socket -> socket splicing. Use the
1163 * piped splicing for that!
1164 */
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001165 i_mode = in->f_path.dentry->d_inode->i_mode;
Jens Axboeb92ce552006-04-11 13:52:07 +02001166 if (unlikely(!S_ISREG(i_mode) && !S_ISBLK(i_mode)))
1167 return -EINVAL;
1168
1169 /*
1170 * neither in nor out is a pipe, setup an internal pipe attached to
1171 * 'out' and transfer the wanted data from 'in' to 'out' through that
1172 */
1173 pipe = current->splice_pipe;
Jens Axboe49570e92006-04-11 13:56:09 +02001174 if (unlikely(!pipe)) {
Jens Axboeb92ce552006-04-11 13:52:07 +02001175 pipe = alloc_pipe_info(NULL);
1176 if (!pipe)
1177 return -ENOMEM;
1178
1179 /*
1180 * We don't have an immediate reader, but we'll read the stuff
Jens Axboe00522fb2006-04-26 14:39:29 +02001181 * out of the pipe right after the splice_to_pipe(). So set
Jens Axboeb92ce552006-04-11 13:52:07 +02001182 * PIPE_READERS appropriately.
1183 */
1184 pipe->readers = 1;
1185
1186 current->splice_pipe = pipe;
1187 }
1188
1189 /*
Ingo Molnar73d62d82006-04-11 13:57:21 +02001190 * Do the splice.
Jens Axboeb92ce552006-04-11 13:52:07 +02001191 */
1192 ret = 0;
1193 bytes = 0;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001194 len = sd->total_len;
1195 flags = sd->flags;
1196
1197 /*
1198 * Don't block on output, we have to drain the direct pipe.
1199 */
1200 sd->flags &= ~SPLICE_F_NONBLOCK;
Jens Axboeb92ce552006-04-11 13:52:07 +02001201
1202 while (len) {
Jens Axboe51a92c02007-07-13 14:11:43 +02001203 size_t read_len;
Tom Zanussia82c53a2008-05-09 13:28:36 +02001204 loff_t pos = sd->pos, prev_pos = pos;
Jens Axboeb92ce552006-04-11 13:52:07 +02001205
Jens Axboebcd4f3a2007-07-16 14:41:49 +02001206 ret = do_splice_to(in, &pos, pipe, len, flags);
Jens Axboe51a92c02007-07-13 14:11:43 +02001207 if (unlikely(ret <= 0))
Jens Axboeb92ce552006-04-11 13:52:07 +02001208 goto out_release;
1209
1210 read_len = ret;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001211 sd->total_len = read_len;
Jens Axboeb92ce552006-04-11 13:52:07 +02001212
1213 /*
1214 * NOTE: nonblocking mode only applies to the input. We
1215 * must not do the output in nonblocking mode as then we
1216 * could get stuck data in the internal pipe:
1217 */
Jens Axboec66ab6f2007-06-12 21:17:17 +02001218 ret = actor(pipe, sd);
Tom Zanussia82c53a2008-05-09 13:28:36 +02001219 if (unlikely(ret <= 0)) {
1220 sd->pos = prev_pos;
Jens Axboeb92ce552006-04-11 13:52:07 +02001221 goto out_release;
Tom Zanussia82c53a2008-05-09 13:28:36 +02001222 }
Jens Axboeb92ce552006-04-11 13:52:07 +02001223
1224 bytes += ret;
1225 len -= ret;
Jens Axboebcd4f3a2007-07-16 14:41:49 +02001226 sd->pos = pos;
Jens Axboeb92ce552006-04-11 13:52:07 +02001227
Tom Zanussia82c53a2008-05-09 13:28:36 +02001228 if (ret < read_len) {
1229 sd->pos = prev_pos + ret;
Jens Axboe51a92c02007-07-13 14:11:43 +02001230 goto out_release;
Tom Zanussia82c53a2008-05-09 13:28:36 +02001231 }
Jens Axboeb92ce552006-04-11 13:52:07 +02001232 }
1233
Jens Axboe9e971982008-01-29 21:05:57 +01001234done:
Jens Axboeb92ce552006-04-11 13:52:07 +02001235 pipe->nrbufs = pipe->curbuf = 0;
Jens Axboe80848702008-01-30 12:24:48 +01001236 file_accessed(in);
Jens Axboeb92ce552006-04-11 13:52:07 +02001237 return bytes;
1238
1239out_release:
1240 /*
1241 * If we did an incomplete transfer we must release
1242 * the pipe buffers in question:
1243 */
Jens Axboe35f3d142010-05-20 10:43:18 +02001244 for (i = 0; i < pipe->buffers; i++) {
Jens Axboeb92ce552006-04-11 13:52:07 +02001245 struct pipe_buffer *buf = pipe->bufs + i;
1246
1247 if (buf->ops) {
1248 buf->ops->release(pipe, buf);
1249 buf->ops = NULL;
1250 }
1251 }
Jens Axboeb92ce552006-04-11 13:52:07 +02001252
Jens Axboe9e971982008-01-29 21:05:57 +01001253 if (!bytes)
1254 bytes = ret;
Jens Axboeb92ce552006-04-11 13:52:07 +02001255
Jens Axboe9e971982008-01-29 21:05:57 +01001256 goto done;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001257}
1258EXPORT_SYMBOL(splice_direct_to_actor);
1259
1260static int direct_splice_actor(struct pipe_inode_info *pipe,
1261 struct splice_desc *sd)
1262{
Jens Axboe6a14b902007-06-14 13:08:55 +02001263 struct file *file = sd->u.file;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001264
Changli Gao2cb4b052010-06-29 13:09:18 +02001265 return do_splice_from(pipe, file, &file->f_pos, sd->total_len,
1266 sd->flags);
Jens Axboec66ab6f2007-06-12 21:17:17 +02001267}
1268
Jens Axboe932cc6d2007-06-21 13:10:21 +02001269/**
1270 * do_splice_direct - splices data directly between two files
1271 * @in: file to splice from
1272 * @ppos: input file offset
1273 * @out: file to splice to
1274 * @len: number of bytes to splice
1275 * @flags: splice modifier flags
1276 *
1277 * Description:
1278 * For use by do_sendfile(). splice can easily emulate sendfile, but
1279 * doing it in the application would incur an extra system call
1280 * (splice in + splice out, as compared to just sendfile()). So this helper
1281 * can splice directly through a process-private pipe.
1282 *
1283 */
Jens Axboec66ab6f2007-06-12 21:17:17 +02001284long do_splice_direct(struct file *in, loff_t *ppos, struct file *out,
1285 size_t len, unsigned int flags)
1286{
1287 struct splice_desc sd = {
1288 .len = len,
1289 .total_len = len,
1290 .flags = flags,
1291 .pos = *ppos,
Jens Axboe6a14b902007-06-14 13:08:55 +02001292 .u.file = out,
Jens Axboec66ab6f2007-06-12 21:17:17 +02001293 };
Jens Axboe51a92c02007-07-13 14:11:43 +02001294 long ret;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001295
1296 ret = splice_direct_to_actor(in, &sd, direct_splice_actor);
Jens Axboe51a92c02007-07-13 14:11:43 +02001297 if (ret > 0)
Tom Zanussia82c53a2008-05-09 13:28:36 +02001298 *ppos = sd.pos;
Jens Axboe51a92c02007-07-13 14:11:43 +02001299
Jens Axboec66ab6f2007-06-12 21:17:17 +02001300 return ret;
Jens Axboeb92ce552006-04-11 13:52:07 +02001301}
1302
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001303static int splice_pipe_to_pipe(struct pipe_inode_info *ipipe,
1304 struct pipe_inode_info *opipe,
1305 size_t len, unsigned int flags);
Jens Axboeddac0d32006-11-04 12:49:32 +01001306
1307/*
Jens Axboe83f91352006-04-02 23:05:09 +02001308 * Determine where to splice to/from.
1309 */
Ingo Molnar529565d2006-04-10 15:18:58 +02001310static long do_splice(struct file *in, loff_t __user *off_in,
1311 struct file *out, loff_t __user *off_out,
1312 size_t len, unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +02001313{
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001314 struct pipe_inode_info *ipipe;
1315 struct pipe_inode_info *opipe;
Jens Axboecbb7e572006-04-11 14:57:50 +02001316 loff_t offset, *off;
Jens Axboea4514eb2006-04-19 15:57:05 +02001317 long ret;
Jens Axboe5274f052006-03-30 15:15:30 +02001318
Linus Torvalds71993e62010-11-28 13:56:09 -08001319 ipipe = get_pipe_info(in);
1320 opipe = get_pipe_info(out);
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001321
1322 if (ipipe && opipe) {
1323 if (off_in || off_out)
1324 return -ESPIPE;
1325
1326 if (!(in->f_mode & FMODE_READ))
1327 return -EBADF;
1328
1329 if (!(out->f_mode & FMODE_WRITE))
1330 return -EBADF;
1331
1332 /* Splicing to self would be fun, but... */
1333 if (ipipe == opipe)
1334 return -EINVAL;
1335
1336 return splice_pipe_to_pipe(ipipe, opipe, len, flags);
1337 }
1338
1339 if (ipipe) {
Ingo Molnar529565d2006-04-10 15:18:58 +02001340 if (off_in)
1341 return -ESPIPE;
Jens Axboeb92ce552006-04-11 13:52:07 +02001342 if (off_out) {
Changli Gao19c9a492010-06-29 13:10:36 +02001343 if (!(out->f_mode & FMODE_PWRITE))
Jens Axboeb92ce552006-04-11 13:52:07 +02001344 return -EINVAL;
Jens Axboecbb7e572006-04-11 14:57:50 +02001345 if (copy_from_user(&offset, off_out, sizeof(loff_t)))
Jens Axboeb92ce552006-04-11 13:52:07 +02001346 return -EFAULT;
Jens Axboecbb7e572006-04-11 14:57:50 +02001347 off = &offset;
1348 } else
1349 off = &out->f_pos;
Ingo Molnar529565d2006-04-10 15:18:58 +02001350
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001351 ret = do_splice_from(ipipe, out, off, len, flags);
Jens Axboea4514eb2006-04-19 15:57:05 +02001352
1353 if (off_out && copy_to_user(off_out, off, sizeof(loff_t)))
1354 ret = -EFAULT;
1355
1356 return ret;
Ingo Molnar529565d2006-04-10 15:18:58 +02001357 }
Jens Axboe5274f052006-03-30 15:15:30 +02001358
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001359 if (opipe) {
Ingo Molnar529565d2006-04-10 15:18:58 +02001360 if (off_out)
1361 return -ESPIPE;
Jens Axboeb92ce552006-04-11 13:52:07 +02001362 if (off_in) {
Changli Gao19c9a492010-06-29 13:10:36 +02001363 if (!(in->f_mode & FMODE_PREAD))
Jens Axboeb92ce552006-04-11 13:52:07 +02001364 return -EINVAL;
Jens Axboecbb7e572006-04-11 14:57:50 +02001365 if (copy_from_user(&offset, off_in, sizeof(loff_t)))
Jens Axboeb92ce552006-04-11 13:52:07 +02001366 return -EFAULT;
Jens Axboecbb7e572006-04-11 14:57:50 +02001367 off = &offset;
1368 } else
1369 off = &in->f_pos;
Ingo Molnar529565d2006-04-10 15:18:58 +02001370
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001371 ret = do_splice_to(in, off, opipe, len, flags);
Jens Axboea4514eb2006-04-19 15:57:05 +02001372
1373 if (off_in && copy_to_user(off_in, off, sizeof(loff_t)))
1374 ret = -EFAULT;
1375
1376 return ret;
Ingo Molnar529565d2006-04-10 15:18:58 +02001377 }
Jens Axboe5274f052006-03-30 15:15:30 +02001378
1379 return -EINVAL;
1380}
1381
Jens Axboe912d35f2006-04-26 10:59:21 +02001382/*
1383 * Map an iov into an array of pages and offset/length tupples. With the
1384 * partial_page structure, we can map several non-contiguous ranges into
1385 * our ones pages[] map instead of splitting that operation into pieces.
1386 * Could easily be exported as a generic helper for other users, in which
1387 * case one would probably want to add a 'max_nr_pages' parameter as well.
1388 */
1389static int get_iovec_page_array(const struct iovec __user *iov,
1390 unsigned int nr_vecs, struct page **pages,
Jens Axboe35f3d142010-05-20 10:43:18 +02001391 struct partial_page *partial, int aligned,
1392 unsigned int pipe_buffers)
Jens Axboe912d35f2006-04-26 10:59:21 +02001393{
1394 int buffers = 0, error = 0;
1395
Jens Axboe912d35f2006-04-26 10:59:21 +02001396 while (nr_vecs) {
1397 unsigned long off, npages;
Linus Torvalds75723952007-10-01 13:17:28 -07001398 struct iovec entry;
Jens Axboe912d35f2006-04-26 10:59:21 +02001399 void __user *base;
1400 size_t len;
1401 int i;
1402
Linus Torvalds75723952007-10-01 13:17:28 -07001403 error = -EFAULT;
Nick Pigginbc40d732008-07-25 19:45:26 -07001404 if (copy_from_user(&entry, iov, sizeof(entry)))
Jens Axboe912d35f2006-04-26 10:59:21 +02001405 break;
Linus Torvalds75723952007-10-01 13:17:28 -07001406
1407 base = entry.iov_base;
1408 len = entry.iov_len;
Jens Axboe912d35f2006-04-26 10:59:21 +02001409
1410 /*
1411 * Sanity check this iovec. 0 read succeeds.
1412 */
Linus Torvalds75723952007-10-01 13:17:28 -07001413 error = 0;
Jens Axboe912d35f2006-04-26 10:59:21 +02001414 if (unlikely(!len))
1415 break;
1416 error = -EFAULT;
Bastian Blank712a30e2008-02-10 16:47:57 +02001417 if (!access_ok(VERIFY_READ, base, len))
Jens Axboe912d35f2006-04-26 10:59:21 +02001418 break;
1419
1420 /*
1421 * Get this base offset and number of pages, then map
1422 * in the user pages.
1423 */
1424 off = (unsigned long) base & ~PAGE_MASK;
Jens Axboe7afa6fd2006-05-01 20:02:33 +02001425
1426 /*
1427 * If asked for alignment, the offset must be zero and the
1428 * length a multiple of the PAGE_SIZE.
1429 */
1430 error = -EINVAL;
1431 if (aligned && (off || len & ~PAGE_MASK))
1432 break;
1433
Jens Axboe912d35f2006-04-26 10:59:21 +02001434 npages = (off + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
Jens Axboe35f3d142010-05-20 10:43:18 +02001435 if (npages > pipe_buffers - buffers)
1436 npages = pipe_buffers - buffers;
Jens Axboe912d35f2006-04-26 10:59:21 +02001437
Nick Pigginbc40d732008-07-25 19:45:26 -07001438 error = get_user_pages_fast((unsigned long)base, npages,
1439 0, &pages[buffers]);
Jens Axboe912d35f2006-04-26 10:59:21 +02001440
1441 if (unlikely(error <= 0))
1442 break;
1443
1444 /*
1445 * Fill this contiguous range into the partial page map.
1446 */
1447 for (i = 0; i < error; i++) {
Jens Axboe75914892006-05-02 12:57:18 +02001448 const int plen = min_t(size_t, len, PAGE_SIZE - off);
Jens Axboe912d35f2006-04-26 10:59:21 +02001449
1450 partial[buffers].offset = off;
1451 partial[buffers].len = plen;
1452
1453 off = 0;
1454 len -= plen;
1455 buffers++;
1456 }
1457
1458 /*
1459 * We didn't complete this iov, stop here since it probably
1460 * means we have to move some of this into a pipe to
1461 * be able to continue.
1462 */
1463 if (len)
1464 break;
1465
1466 /*
1467 * Don't continue if we mapped fewer pages than we asked for,
1468 * or if we mapped the max number of pages that we have
1469 * room for.
1470 */
Jens Axboe35f3d142010-05-20 10:43:18 +02001471 if (error < npages || buffers == pipe_buffers)
Jens Axboe912d35f2006-04-26 10:59:21 +02001472 break;
1473
1474 nr_vecs--;
1475 iov++;
1476 }
1477
Jens Axboe912d35f2006-04-26 10:59:21 +02001478 if (buffers)
1479 return buffers;
1480
1481 return error;
1482}
1483
Jens Axboe6a14b902007-06-14 13:08:55 +02001484static int pipe_to_user(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
1485 struct splice_desc *sd)
1486{
1487 char *src;
1488 int ret;
1489
Jens Axboe6a14b902007-06-14 13:08:55 +02001490 /*
1491 * See if we can use the atomic maps, by prefaulting in the
1492 * pages and doing an atomic copy
1493 */
1494 if (!fault_in_pages_writeable(sd->u.userptr, sd->len)) {
1495 src = buf->ops->map(pipe, buf, 1);
1496 ret = __copy_to_user_inatomic(sd->u.userptr, src + buf->offset,
1497 sd->len);
1498 buf->ops->unmap(pipe, buf, src);
1499 if (!ret) {
1500 ret = sd->len;
1501 goto out;
1502 }
1503 }
1504
1505 /*
1506 * No dice, use slow non-atomic map and copy
1507 */
1508 src = buf->ops->map(pipe, buf, 0);
1509
1510 ret = sd->len;
1511 if (copy_to_user(sd->u.userptr, src + buf->offset, sd->len))
1512 ret = -EFAULT;
1513
Jens Axboe6866bef2007-10-16 10:01:29 +02001514 buf->ops->unmap(pipe, buf, src);
Jens Axboe6a14b902007-06-14 13:08:55 +02001515out:
1516 if (ret > 0)
1517 sd->u.userptr += ret;
Jens Axboe6a14b902007-06-14 13:08:55 +02001518 return ret;
1519}
1520
1521/*
1522 * For lack of a better implementation, implement vmsplice() to userspace
1523 * as a simple copy of the pipes pages to the user iov.
1524 */
1525static long vmsplice_to_user(struct file *file, const struct iovec __user *iov,
1526 unsigned long nr_segs, unsigned int flags)
1527{
1528 struct pipe_inode_info *pipe;
1529 struct splice_desc sd;
1530 ssize_t size;
1531 int error;
1532 long ret;
1533
Linus Torvalds71993e62010-11-28 13:56:09 -08001534 pipe = get_pipe_info(file);
Jens Axboe6a14b902007-06-14 13:08:55 +02001535 if (!pipe)
1536 return -EBADF;
1537
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001538 pipe_lock(pipe);
Jens Axboe6a14b902007-06-14 13:08:55 +02001539
1540 error = ret = 0;
1541 while (nr_segs) {
1542 void __user *base;
1543 size_t len;
1544
1545 /*
1546 * Get user address base and length for this iovec.
1547 */
1548 error = get_user(base, &iov->iov_base);
1549 if (unlikely(error))
1550 break;
1551 error = get_user(len, &iov->iov_len);
1552 if (unlikely(error))
1553 break;
1554
1555 /*
1556 * Sanity check this iovec. 0 read succeeds.
1557 */
1558 if (unlikely(!len))
1559 break;
1560 if (unlikely(!base)) {
1561 error = -EFAULT;
1562 break;
1563 }
1564
Jens Axboe88119302008-02-08 08:49:14 -08001565 if (unlikely(!access_ok(VERIFY_WRITE, base, len))) {
1566 error = -EFAULT;
1567 break;
1568 }
1569
Jens Axboe6a14b902007-06-14 13:08:55 +02001570 sd.len = 0;
1571 sd.total_len = len;
1572 sd.flags = flags;
1573 sd.u.userptr = base;
1574 sd.pos = 0;
1575
1576 size = __splice_from_pipe(pipe, &sd, pipe_to_user);
1577 if (size < 0) {
1578 if (!ret)
1579 ret = size;
1580
1581 break;
1582 }
1583
1584 ret += size;
1585
1586 if (size < len)
1587 break;
1588
1589 nr_segs--;
1590 iov++;
1591 }
1592
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001593 pipe_unlock(pipe);
Jens Axboe6a14b902007-06-14 13:08:55 +02001594
1595 if (!ret)
1596 ret = error;
1597
1598 return ret;
1599}
1600
Jens Axboe912d35f2006-04-26 10:59:21 +02001601/*
1602 * vmsplice splices a user address range into a pipe. It can be thought of
1603 * as splice-from-memory, where the regular splice is splice-from-file (or
1604 * to file). In both cases the output is a pipe, naturally.
Jens Axboe912d35f2006-04-26 10:59:21 +02001605 */
Jens Axboe6a14b902007-06-14 13:08:55 +02001606static long vmsplice_to_pipe(struct file *file, const struct iovec __user *iov,
1607 unsigned long nr_segs, unsigned int flags)
Jens Axboe912d35f2006-04-26 10:59:21 +02001608{
Jens Axboeddac0d32006-11-04 12:49:32 +01001609 struct pipe_inode_info *pipe;
Jens Axboe35f3d142010-05-20 10:43:18 +02001610 struct page *pages[PIPE_DEF_BUFFERS];
1611 struct partial_page partial[PIPE_DEF_BUFFERS];
Jens Axboe912d35f2006-04-26 10:59:21 +02001612 struct splice_pipe_desc spd = {
1613 .pages = pages,
1614 .partial = partial,
1615 .flags = flags,
1616 .ops = &user_page_pipe_buf_ops,
Jens Axboebbdfc2f2007-11-06 23:29:47 -08001617 .spd_release = spd_release_page,
Jens Axboe912d35f2006-04-26 10:59:21 +02001618 };
Jens Axboe35f3d142010-05-20 10:43:18 +02001619 long ret;
Jens Axboe912d35f2006-04-26 10:59:21 +02001620
Linus Torvalds71993e62010-11-28 13:56:09 -08001621 pipe = get_pipe_info(file);
Jens Axboeddac0d32006-11-04 12:49:32 +01001622 if (!pipe)
Jens Axboe912d35f2006-04-26 10:59:21 +02001623 return -EBADF;
Jens Axboe912d35f2006-04-26 10:59:21 +02001624
Jens Axboe35f3d142010-05-20 10:43:18 +02001625 if (splice_grow_spd(pipe, &spd))
1626 return -ENOMEM;
Jens Axboe912d35f2006-04-26 10:59:21 +02001627
Jens Axboe35f3d142010-05-20 10:43:18 +02001628 spd.nr_pages = get_iovec_page_array(iov, nr_segs, spd.pages,
1629 spd.partial, flags & SPLICE_F_GIFT,
1630 pipe->buffers);
1631 if (spd.nr_pages <= 0)
1632 ret = spd.nr_pages;
1633 else
1634 ret = splice_to_pipe(pipe, &spd);
1635
1636 splice_shrink_spd(pipe, &spd);
1637 return ret;
Jens Axboe912d35f2006-04-26 10:59:21 +02001638}
1639
Jens Axboe6a14b902007-06-14 13:08:55 +02001640/*
1641 * Note that vmsplice only really supports true splicing _from_ user memory
1642 * to a pipe, not the other way around. Splicing from user memory is a simple
1643 * operation that can be supported without any funky alignment restrictions
1644 * or nasty vm tricks. We simply map in the user memory and fill them into
1645 * a pipe. The reverse isn't quite as easy, though. There are two possible
1646 * solutions for that:
1647 *
1648 * - memcpy() the data internally, at which point we might as well just
1649 * do a regular read() on the buffer anyway.
1650 * - Lots of nasty vm tricks, that are neither fast nor flexible (it
1651 * has restriction limitations on both ends of the pipe).
1652 *
1653 * Currently we punt and implement it as a normal copy, see pipe_to_user().
1654 *
1655 */
Heiko Carstens836f92a2009-01-14 14:14:33 +01001656SYSCALL_DEFINE4(vmsplice, int, fd, const struct iovec __user *, iov,
1657 unsigned long, nr_segs, unsigned int, flags)
Jens Axboe912d35f2006-04-26 10:59:21 +02001658{
1659 struct file *file;
1660 long error;
1661 int fput;
1662
Jens Axboe6a14b902007-06-14 13:08:55 +02001663 if (unlikely(nr_segs > UIO_MAXIOV))
1664 return -EINVAL;
1665 else if (unlikely(!nr_segs))
1666 return 0;
1667
Jens Axboe912d35f2006-04-26 10:59:21 +02001668 error = -EBADF;
1669 file = fget_light(fd, &fput);
1670 if (file) {
1671 if (file->f_mode & FMODE_WRITE)
Jens Axboe6a14b902007-06-14 13:08:55 +02001672 error = vmsplice_to_pipe(file, iov, nr_segs, flags);
1673 else if (file->f_mode & FMODE_READ)
1674 error = vmsplice_to_user(file, iov, nr_segs, flags);
Jens Axboe912d35f2006-04-26 10:59:21 +02001675
1676 fput_light(file, fput);
1677 }
1678
1679 return error;
1680}
1681
Heiko Carstens836f92a2009-01-14 14:14:33 +01001682SYSCALL_DEFINE6(splice, int, fd_in, loff_t __user *, off_in,
1683 int, fd_out, loff_t __user *, off_out,
1684 size_t, len, unsigned int, flags)
Jens Axboe5274f052006-03-30 15:15:30 +02001685{
1686 long error;
1687 struct file *in, *out;
1688 int fput_in, fput_out;
1689
1690 if (unlikely(!len))
1691 return 0;
1692
1693 error = -EBADF;
Ingo Molnar529565d2006-04-10 15:18:58 +02001694 in = fget_light(fd_in, &fput_in);
Jens Axboe5274f052006-03-30 15:15:30 +02001695 if (in) {
1696 if (in->f_mode & FMODE_READ) {
Ingo Molnar529565d2006-04-10 15:18:58 +02001697 out = fget_light(fd_out, &fput_out);
Jens Axboe5274f052006-03-30 15:15:30 +02001698 if (out) {
1699 if (out->f_mode & FMODE_WRITE)
Ingo Molnar529565d2006-04-10 15:18:58 +02001700 error = do_splice(in, off_in,
1701 out, off_out,
1702 len, flags);
Jens Axboe5274f052006-03-30 15:15:30 +02001703 fput_light(out, fput_out);
1704 }
1705 }
1706
1707 fput_light(in, fput_in);
1708 }
1709
1710 return error;
1711}
Jens Axboe70524492006-04-11 15:51:17 +02001712
1713/*
Jens Axboeaadd06e2006-07-10 11:00:01 +02001714 * Make sure there's data to read. Wait for input if we can, otherwise
1715 * return an appropriate error.
1716 */
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001717static int ipipe_prep(struct pipe_inode_info *pipe, unsigned int flags)
Jens Axboeaadd06e2006-07-10 11:00:01 +02001718{
1719 int ret;
1720
1721 /*
1722 * Check ->nrbufs without the inode lock first. This function
1723 * is speculative anyways, so missing one is ok.
1724 */
1725 if (pipe->nrbufs)
1726 return 0;
1727
1728 ret = 0;
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001729 pipe_lock(pipe);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001730
1731 while (!pipe->nrbufs) {
1732 if (signal_pending(current)) {
1733 ret = -ERESTARTSYS;
1734 break;
1735 }
1736 if (!pipe->writers)
1737 break;
1738 if (!pipe->waiting_writers) {
1739 if (flags & SPLICE_F_NONBLOCK) {
1740 ret = -EAGAIN;
1741 break;
1742 }
1743 }
1744 pipe_wait(pipe);
1745 }
1746
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001747 pipe_unlock(pipe);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001748 return ret;
1749}
1750
1751/*
1752 * Make sure there's writeable room. Wait for room if we can, otherwise
1753 * return an appropriate error.
1754 */
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001755static int opipe_prep(struct pipe_inode_info *pipe, unsigned int flags)
Jens Axboeaadd06e2006-07-10 11:00:01 +02001756{
1757 int ret;
1758
1759 /*
1760 * Check ->nrbufs without the inode lock first. This function
1761 * is speculative anyways, so missing one is ok.
1762 */
Jens Axboe35f3d142010-05-20 10:43:18 +02001763 if (pipe->nrbufs < pipe->buffers)
Jens Axboeaadd06e2006-07-10 11:00:01 +02001764 return 0;
1765
1766 ret = 0;
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001767 pipe_lock(pipe);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001768
Jens Axboe35f3d142010-05-20 10:43:18 +02001769 while (pipe->nrbufs >= pipe->buffers) {
Jens Axboeaadd06e2006-07-10 11:00:01 +02001770 if (!pipe->readers) {
1771 send_sig(SIGPIPE, current, 0);
1772 ret = -EPIPE;
1773 break;
1774 }
1775 if (flags & SPLICE_F_NONBLOCK) {
1776 ret = -EAGAIN;
1777 break;
1778 }
1779 if (signal_pending(current)) {
1780 ret = -ERESTARTSYS;
1781 break;
1782 }
1783 pipe->waiting_writers++;
1784 pipe_wait(pipe);
1785 pipe->waiting_writers--;
1786 }
1787
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001788 pipe_unlock(pipe);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001789 return ret;
1790}
1791
1792/*
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001793 * Splice contents of ipipe to opipe.
1794 */
1795static int splice_pipe_to_pipe(struct pipe_inode_info *ipipe,
1796 struct pipe_inode_info *opipe,
1797 size_t len, unsigned int flags)
1798{
1799 struct pipe_buffer *ibuf, *obuf;
1800 int ret = 0, nbuf;
1801 bool input_wakeup = false;
1802
1803
1804retry:
1805 ret = ipipe_prep(ipipe, flags);
1806 if (ret)
1807 return ret;
1808
1809 ret = opipe_prep(opipe, flags);
1810 if (ret)
1811 return ret;
1812
1813 /*
1814 * Potential ABBA deadlock, work around it by ordering lock
1815 * grabbing by pipe info address. Otherwise two different processes
1816 * could deadlock (one doing tee from A -> B, the other from B -> A).
1817 */
1818 pipe_double_lock(ipipe, opipe);
1819
1820 do {
1821 if (!opipe->readers) {
1822 send_sig(SIGPIPE, current, 0);
1823 if (!ret)
1824 ret = -EPIPE;
1825 break;
1826 }
1827
1828 if (!ipipe->nrbufs && !ipipe->writers)
1829 break;
1830
1831 /*
1832 * Cannot make any progress, because either the input
1833 * pipe is empty or the output pipe is full.
1834 */
Jens Axboe35f3d142010-05-20 10:43:18 +02001835 if (!ipipe->nrbufs || opipe->nrbufs >= opipe->buffers) {
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001836 /* Already processed some buffers, break */
1837 if (ret)
1838 break;
1839
1840 if (flags & SPLICE_F_NONBLOCK) {
1841 ret = -EAGAIN;
1842 break;
1843 }
1844
1845 /*
1846 * We raced with another reader/writer and haven't
1847 * managed to process any buffers. A zero return
1848 * value means EOF, so retry instead.
1849 */
1850 pipe_unlock(ipipe);
1851 pipe_unlock(opipe);
1852 goto retry;
1853 }
1854
1855 ibuf = ipipe->bufs + ipipe->curbuf;
Jens Axboe35f3d142010-05-20 10:43:18 +02001856 nbuf = (opipe->curbuf + opipe->nrbufs) & (opipe->buffers - 1);
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001857 obuf = opipe->bufs + nbuf;
1858
1859 if (len >= ibuf->len) {
1860 /*
1861 * Simply move the whole buffer from ipipe to opipe
1862 */
1863 *obuf = *ibuf;
1864 ibuf->ops = NULL;
1865 opipe->nrbufs++;
Jens Axboe35f3d142010-05-20 10:43:18 +02001866 ipipe->curbuf = (ipipe->curbuf + 1) & (ipipe->buffers - 1);
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001867 ipipe->nrbufs--;
1868 input_wakeup = true;
1869 } else {
1870 /*
1871 * Get a reference to this pipe buffer,
1872 * so we can copy the contents over.
1873 */
1874 ibuf->ops->get(ipipe, ibuf);
1875 *obuf = *ibuf;
1876
1877 /*
1878 * Don't inherit the gift flag, we need to
1879 * prevent multiple steals of this page.
1880 */
1881 obuf->flags &= ~PIPE_BUF_FLAG_GIFT;
1882
1883 obuf->len = len;
1884 opipe->nrbufs++;
1885 ibuf->offset += obuf->len;
1886 ibuf->len -= obuf->len;
1887 }
1888 ret += obuf->len;
1889 len -= obuf->len;
1890 } while (len);
1891
1892 pipe_unlock(ipipe);
1893 pipe_unlock(opipe);
1894
1895 /*
1896 * If we put data in the output pipe, wakeup any potential readers.
1897 */
Namhyung Kim825cdcb2011-05-23 19:58:53 +02001898 if (ret > 0)
1899 wakeup_pipe_readers(opipe);
1900
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001901 if (input_wakeup)
1902 wakeup_pipe_writers(ipipe);
1903
1904 return ret;
1905}
1906
1907/*
Jens Axboe70524492006-04-11 15:51:17 +02001908 * Link contents of ipipe to opipe.
1909 */
1910static int link_pipe(struct pipe_inode_info *ipipe,
1911 struct pipe_inode_info *opipe,
1912 size_t len, unsigned int flags)
1913{
1914 struct pipe_buffer *ibuf, *obuf;
Jens Axboeaadd06e2006-07-10 11:00:01 +02001915 int ret = 0, i = 0, nbuf;
Jens Axboe70524492006-04-11 15:51:17 +02001916
1917 /*
1918 * Potential ABBA deadlock, work around it by ordering lock
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001919 * grabbing by pipe info address. Otherwise two different processes
Jens Axboe70524492006-04-11 15:51:17 +02001920 * could deadlock (one doing tee from A -> B, the other from B -> A).
1921 */
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001922 pipe_double_lock(ipipe, opipe);
Jens Axboe70524492006-04-11 15:51:17 +02001923
Jens Axboeaadd06e2006-07-10 11:00:01 +02001924 do {
Jens Axboe70524492006-04-11 15:51:17 +02001925 if (!opipe->readers) {
1926 send_sig(SIGPIPE, current, 0);
1927 if (!ret)
1928 ret = -EPIPE;
1929 break;
1930 }
Jens Axboe70524492006-04-11 15:51:17 +02001931
1932 /*
Jens Axboeaadd06e2006-07-10 11:00:01 +02001933 * If we have iterated all input buffers or ran out of
1934 * output room, break.
Jens Axboe70524492006-04-11 15:51:17 +02001935 */
Jens Axboe35f3d142010-05-20 10:43:18 +02001936 if (i >= ipipe->nrbufs || opipe->nrbufs >= opipe->buffers)
Jens Axboe70524492006-04-11 15:51:17 +02001937 break;
Jens Axboeaadd06e2006-07-10 11:00:01 +02001938
Jens Axboe35f3d142010-05-20 10:43:18 +02001939 ibuf = ipipe->bufs + ((ipipe->curbuf + i) & (ipipe->buffers-1));
1940 nbuf = (opipe->curbuf + opipe->nrbufs) & (opipe->buffers - 1);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001941
Jens Axboe2a272502006-04-19 15:56:40 +02001942 /*
Jens Axboeaadd06e2006-07-10 11:00:01 +02001943 * Get a reference to this pipe buffer,
1944 * so we can copy the contents over.
Jens Axboe2a272502006-04-19 15:56:40 +02001945 */
Jens Axboeaadd06e2006-07-10 11:00:01 +02001946 ibuf->ops->get(ipipe, ibuf);
Jens Axboe70524492006-04-11 15:51:17 +02001947
Jens Axboeaadd06e2006-07-10 11:00:01 +02001948 obuf = opipe->bufs + nbuf;
1949 *obuf = *ibuf;
Jens Axboe70524492006-04-11 15:51:17 +02001950
Jens Axboeaadd06e2006-07-10 11:00:01 +02001951 /*
1952 * Don't inherit the gift flag, we need to
1953 * prevent multiple steals of this page.
1954 */
1955 obuf->flags &= ~PIPE_BUF_FLAG_GIFT;
1956
1957 if (obuf->len > len)
1958 obuf->len = len;
1959
1960 opipe->nrbufs++;
1961 ret += obuf->len;
1962 len -= obuf->len;
1963 i++;
1964 } while (len);
Jens Axboe70524492006-04-11 15:51:17 +02001965
Jens Axboe02cf01a2008-02-20 10:34:51 +01001966 /*
1967 * return EAGAIN if we have the potential of some data in the
1968 * future, otherwise just return 0
1969 */
1970 if (!ret && ipipe->waiting_writers && (flags & SPLICE_F_NONBLOCK))
1971 ret = -EAGAIN;
1972
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001973 pipe_unlock(ipipe);
1974 pipe_unlock(opipe);
Jens Axboe70524492006-04-11 15:51:17 +02001975
Jens Axboeaadd06e2006-07-10 11:00:01 +02001976 /*
1977 * If we put data in the output pipe, wakeup any potential readers.
1978 */
Namhyung Kim825cdcb2011-05-23 19:58:53 +02001979 if (ret > 0)
1980 wakeup_pipe_readers(opipe);
Jens Axboe70524492006-04-11 15:51:17 +02001981
1982 return ret;
1983}
1984
1985/*
1986 * This is a tee(1) implementation that works on pipes. It doesn't copy
1987 * any data, it simply references the 'in' pages on the 'out' pipe.
1988 * The 'flags' used are the SPLICE_F_* variants, currently the only
1989 * applicable one is SPLICE_F_NONBLOCK.
1990 */
1991static long do_tee(struct file *in, struct file *out, size_t len,
1992 unsigned int flags)
1993{
Linus Torvalds71993e62010-11-28 13:56:09 -08001994 struct pipe_inode_info *ipipe = get_pipe_info(in);
1995 struct pipe_inode_info *opipe = get_pipe_info(out);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001996 int ret = -EINVAL;
Jens Axboe70524492006-04-11 15:51:17 +02001997
1998 /*
Jens Axboeaadd06e2006-07-10 11:00:01 +02001999 * Duplicate the contents of ipipe to opipe without actually
2000 * copying the data.
Jens Axboe70524492006-04-11 15:51:17 +02002001 */
Jens Axboeaadd06e2006-07-10 11:00:01 +02002002 if (ipipe && opipe && ipipe != opipe) {
2003 /*
2004 * Keep going, unless we encounter an error. The ipipe/opipe
2005 * ordering doesn't really matter.
2006 */
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02002007 ret = ipipe_prep(ipipe, flags);
Jens Axboeaadd06e2006-07-10 11:00:01 +02002008 if (!ret) {
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02002009 ret = opipe_prep(opipe, flags);
Jens Axboe02cf01a2008-02-20 10:34:51 +01002010 if (!ret)
Jens Axboeaadd06e2006-07-10 11:00:01 +02002011 ret = link_pipe(ipipe, opipe, len, flags);
Jens Axboeaadd06e2006-07-10 11:00:01 +02002012 }
2013 }
Jens Axboe70524492006-04-11 15:51:17 +02002014
Jens Axboeaadd06e2006-07-10 11:00:01 +02002015 return ret;
Jens Axboe70524492006-04-11 15:51:17 +02002016}
2017
Heiko Carstens836f92a2009-01-14 14:14:33 +01002018SYSCALL_DEFINE4(tee, int, fdin, int, fdout, size_t, len, unsigned int, flags)
Jens Axboe70524492006-04-11 15:51:17 +02002019{
2020 struct file *in;
2021 int error, fput_in;
2022
2023 if (unlikely(!len))
2024 return 0;
2025
2026 error = -EBADF;
2027 in = fget_light(fdin, &fput_in);
2028 if (in) {
2029 if (in->f_mode & FMODE_READ) {
2030 int fput_out;
2031 struct file *out = fget_light(fdout, &fput_out);
2032
2033 if (out) {
2034 if (out->f_mode & FMODE_WRITE)
2035 error = do_tee(in, out, len, flags);
2036 fput_light(out, fput_out);
2037 }
2038 }
2039 fput_light(in, fput_in);
2040 }
2041
2042 return error;
2043}