blob: 50a5d978da1698f68cf8d2d2eca97474887ea223 [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 Axboecac36bb02007-06-14 13:10:48 +020097static int page_cache_pipe_buf_confirm(struct pipe_inode_info *pipe,
98 struct pipe_buffer *buf)
Jens Axboe5274f052006-03-30 15:15:30 +020099{
100 struct page *page = buf->page;
Jens Axboe49d0b212006-04-10 09:04:41 +0200101 int err;
Jens Axboe5274f052006-03-30 15:15:30 +0200102
103 if (!PageUptodate(page)) {
Jens Axboe49d0b212006-04-10 09:04:41 +0200104 lock_page(page);
105
106 /*
107 * Page got truncated/unhashed. This will cause a 0-byte
Ingo Molnar73d62d82006-04-11 13:57:21 +0200108 * splice, if this is the first page.
Jens Axboe49d0b212006-04-10 09:04:41 +0200109 */
110 if (!page->mapping) {
111 err = -ENODATA;
112 goto error;
113 }
114
115 /*
Ingo Molnar73d62d82006-04-11 13:57:21 +0200116 * Uh oh, read-error from disk.
Jens Axboe49d0b212006-04-10 09:04:41 +0200117 */
118 if (!PageUptodate(page)) {
119 err = -EIO;
120 goto error;
121 }
122
123 /*
Jens Axboef84d7512006-05-01 19:59:03 +0200124 * Page is ok afterall, we are done.
Jens Axboe49d0b212006-04-10 09:04:41 +0200125 */
Jens Axboe5274f052006-03-30 15:15:30 +0200126 unlock_page(page);
Jens Axboe5274f052006-03-30 15:15:30 +0200127 }
128
Jens Axboef84d7512006-05-01 19:59:03 +0200129 return 0;
Jens Axboe49d0b212006-04-10 09:04:41 +0200130error:
131 unlock_page(page);
Jens Axboef84d7512006-05-01 19:59:03 +0200132 return err;
Jens Axboe70524492006-04-11 15:51:17 +0200133}
134
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 Axboecac36bb02007-06-14 13:10:48 +0200139 .confirm = page_cache_pipe_buf_confirm,
Jens Axboe5274f052006-03-30 15:15:30 +0200140 .release = page_cache_pipe_buf_release,
Jens Axboe5abc97a2006-03-30 15:16:46 +0200141 .steal = page_cache_pipe_buf_steal,
Jens Axboef84d7512006-05-01 19:59:03 +0200142 .get = generic_pipe_buf_get,
Jens Axboe5274f052006-03-30 15:15:30 +0200143};
144
Jens Axboe912d35f2006-04-26 10:59:21 +0200145static int user_page_pipe_buf_steal(struct pipe_inode_info *pipe,
146 struct pipe_buffer *buf)
147{
Jens Axboe7afa6fd2006-05-01 20:02:33 +0200148 if (!(buf->flags & PIPE_BUF_FLAG_GIFT))
149 return 1;
150
Jens Axboe14328732006-05-03 10:35:26 +0200151 buf->flags |= PIPE_BUF_FLAG_LRU;
Jens Axboe330ab712006-05-02 15:29:57 +0200152 return generic_pipe_buf_steal(pipe, buf);
Jens Axboe912d35f2006-04-26 10:59:21 +0200153}
154
Eric Dumazetd4c3cca2006-12-13 00:34:04 -0800155static const struct pipe_buf_operations user_page_pipe_buf_ops = {
Jens Axboe912d35f2006-04-26 10:59:21 +0200156 .can_merge = 0,
Jens Axboef84d7512006-05-01 19:59:03 +0200157 .map = generic_pipe_buf_map,
158 .unmap = generic_pipe_buf_unmap,
Jens Axboecac36bb02007-06-14 13:10:48 +0200159 .confirm = generic_pipe_buf_confirm,
Jens Axboe912d35f2006-04-26 10:59:21 +0200160 .release = page_cache_pipe_buf_release,
161 .steal = user_page_pipe_buf_steal,
Jens Axboef84d7512006-05-01 19:59:03 +0200162 .get = generic_pipe_buf_get,
Jens Axboe912d35f2006-04-26 10:59:21 +0200163};
164
Jens Axboe932cc6d2007-06-21 13:10:21 +0200165/**
166 * splice_to_pipe - fill passed data into a pipe
167 * @pipe: pipe to fill
168 * @spd: data to fill
169 *
170 * Description:
Randy Dunlap79685b82007-07-27 08:08:51 +0200171 * @spd contains a map of pages and len/offset tuples, along with
Jens Axboe932cc6d2007-06-21 13:10:21 +0200172 * the struct pipe_buf_operations associated with these pages. This
173 * function will link that data to the pipe.
174 *
Jens Axboe83f91352006-04-02 23:05:09 +0200175 */
Jens Axboed6b29d72007-06-04 09:59:47 +0200176ssize_t splice_to_pipe(struct pipe_inode_info *pipe,
177 struct splice_pipe_desc *spd)
Jens Axboe5274f052006-03-30 15:15:30 +0200178{
Jens Axboe00de00b2007-06-15 13:14:22 +0200179 unsigned int spd_pages = spd->nr_pages;
Jens Axboe912d35f2006-04-26 10:59:21 +0200180 int ret, do_wakeup, page_nr;
Jens Axboe5274f052006-03-30 15:15:30 +0200181
182 ret = 0;
183 do_wakeup = 0;
Jens Axboe912d35f2006-04-26 10:59:21 +0200184 page_nr = 0;
Jens Axboe5274f052006-03-30 15:15:30 +0200185
Miklos Szeredi61e0d472009-04-14 19:48:41 +0200186 pipe_lock(pipe);
Jens Axboe5274f052006-03-30 15:15:30 +0200187
Jens Axboe5274f052006-03-30 15:15:30 +0200188 for (;;) {
Ingo Molnar3a326a22006-04-10 15:18:35 +0200189 if (!pipe->readers) {
Jens Axboe5274f052006-03-30 15:15:30 +0200190 send_sig(SIGPIPE, current, 0);
191 if (!ret)
192 ret = -EPIPE;
193 break;
194 }
195
Jens Axboe35f3d142010-05-20 10:43:18 +0200196 if (pipe->nrbufs < pipe->buffers) {
197 int newbuf = (pipe->curbuf + pipe->nrbufs) & (pipe->buffers - 1);
Ingo Molnar3a326a22006-04-10 15:18:35 +0200198 struct pipe_buffer *buf = pipe->bufs + newbuf;
Jens Axboe5274f052006-03-30 15:15:30 +0200199
Jens Axboe912d35f2006-04-26 10:59:21 +0200200 buf->page = spd->pages[page_nr];
201 buf->offset = spd->partial[page_nr].offset;
202 buf->len = spd->partial[page_nr].len;
Jens Axboe497f9622007-06-11 12:00:45 +0200203 buf->private = spd->partial[page_nr].private;
Jens Axboe912d35f2006-04-26 10:59:21 +0200204 buf->ops = spd->ops;
Jens Axboe7afa6fd2006-05-01 20:02:33 +0200205 if (spd->flags & SPLICE_F_GIFT)
206 buf->flags |= PIPE_BUF_FLAG_GIFT;
207
Jens Axboe6f767b02006-04-11 13:53:56 +0200208 pipe->nrbufs++;
Jens Axboe912d35f2006-04-26 10:59:21 +0200209 page_nr++;
210 ret += buf->len;
211
Jens Axboe6f767b02006-04-11 13:53:56 +0200212 if (pipe->inode)
213 do_wakeup = 1;
Jens Axboe5274f052006-03-30 15:15:30 +0200214
Jens Axboe912d35f2006-04-26 10:59:21 +0200215 if (!--spd->nr_pages)
Jens Axboe5274f052006-03-30 15:15:30 +0200216 break;
Jens Axboe35f3d142010-05-20 10:43:18 +0200217 if (pipe->nrbufs < pipe->buffers)
Jens Axboe5274f052006-03-30 15:15:30 +0200218 continue;
219
220 break;
221 }
222
Jens Axboe912d35f2006-04-26 10:59:21 +0200223 if (spd->flags & SPLICE_F_NONBLOCK) {
Linus Torvalds29e35092006-04-02 12:46:35 -0700224 if (!ret)
225 ret = -EAGAIN;
226 break;
227 }
228
Jens Axboe5274f052006-03-30 15:15:30 +0200229 if (signal_pending(current)) {
230 if (!ret)
231 ret = -ERESTARTSYS;
232 break;
233 }
234
235 if (do_wakeup) {
Jens Axboec0bd1f62006-04-10 09:03:32 +0200236 smp_mb();
Ingo Molnar3a326a22006-04-10 15:18:35 +0200237 if (waitqueue_active(&pipe->wait))
238 wake_up_interruptible_sync(&pipe->wait);
239 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
Jens Axboe5274f052006-03-30 15:15:30 +0200240 do_wakeup = 0;
241 }
242
Ingo Molnar3a326a22006-04-10 15:18:35 +0200243 pipe->waiting_writers++;
244 pipe_wait(pipe);
245 pipe->waiting_writers--;
Jens Axboe5274f052006-03-30 15:15:30 +0200246 }
247
Miklos Szeredi61e0d472009-04-14 19:48:41 +0200248 pipe_unlock(pipe);
Jens Axboe5274f052006-03-30 15:15:30 +0200249
Miklos Szeredi61e0d472009-04-14 19:48:41 +0200250 if (do_wakeup) {
251 smp_mb();
252 if (waitqueue_active(&pipe->wait))
253 wake_up_interruptible(&pipe->wait);
254 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
Jens Axboe5274f052006-03-30 15:15:30 +0200255 }
256
Jens Axboe00de00b2007-06-15 13:14:22 +0200257 while (page_nr < spd_pages)
Jens Axboebbdfc2f2007-11-06 23:29:47 -0800258 spd->spd_release(spd, page_nr++);
Jens Axboe5274f052006-03-30 15:15:30 +0200259
260 return ret;
261}
262
Jens Axboebbdfc2f2007-11-06 23:29:47 -0800263static void spd_release_page(struct splice_pipe_desc *spd, unsigned int i)
264{
265 page_cache_release(spd->pages[i]);
266}
267
Jens Axboe35f3d142010-05-20 10:43:18 +0200268/*
269 * Check if we need to grow the arrays holding pages and partial page
270 * descriptions.
271 */
272int splice_grow_spd(struct pipe_inode_info *pipe, struct splice_pipe_desc *spd)
273{
274 if (pipe->buffers <= PIPE_DEF_BUFFERS)
275 return 0;
276
277 spd->pages = kmalloc(pipe->buffers * sizeof(struct page *), GFP_KERNEL);
278 spd->partial = kmalloc(pipe->buffers * sizeof(struct partial_page), GFP_KERNEL);
279
280 if (spd->pages && spd->partial)
281 return 0;
282
283 kfree(spd->pages);
284 kfree(spd->partial);
285 return -ENOMEM;
286}
287
288void splice_shrink_spd(struct pipe_inode_info *pipe,
289 struct splice_pipe_desc *spd)
290{
291 if (pipe->buffers <= PIPE_DEF_BUFFERS)
292 return;
293
294 kfree(spd->pages);
295 kfree(spd->partial);
296}
297
Ingo Molnar3a326a22006-04-10 15:18:35 +0200298static int
Jens Axboecbb7e572006-04-11 14:57:50 +0200299__generic_file_splice_read(struct file *in, loff_t *ppos,
300 struct pipe_inode_info *pipe, size_t len,
301 unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +0200302{
303 struct address_space *mapping = in->f_mapping;
Fengguang Wud8983912007-07-19 01:48:06 -0700304 unsigned int loff, nr_pages, req_pages;
Jens Axboe35f3d142010-05-20 10:43:18 +0200305 struct page *pages[PIPE_DEF_BUFFERS];
306 struct partial_page partial[PIPE_DEF_BUFFERS];
Jens Axboe5274f052006-03-30 15:15:30 +0200307 struct page *page;
Jens Axboe91ad66e2006-04-19 15:55:10 +0200308 pgoff_t index, end_index;
309 loff_t isize;
Jens Axboeeb207962006-04-27 11:05:22 +0200310 int error, page_nr;
Jens Axboe912d35f2006-04-26 10:59:21 +0200311 struct splice_pipe_desc spd = {
312 .pages = pages,
313 .partial = partial,
314 .flags = flags,
315 .ops = &page_cache_pipe_buf_ops,
Jens Axboebbdfc2f2007-11-06 23:29:47 -0800316 .spd_release = spd_release_page,
Jens Axboe912d35f2006-04-26 10:59:21 +0200317 };
Jens Axboe5274f052006-03-30 15:15:30 +0200318
Jens Axboe35f3d142010-05-20 10:43:18 +0200319 if (splice_grow_spd(pipe, &spd))
320 return -ENOMEM;
321
Jens Axboecbb7e572006-04-11 14:57:50 +0200322 index = *ppos >> PAGE_CACHE_SHIFT;
Jens Axboe912d35f2006-04-26 10:59:21 +0200323 loff = *ppos & ~PAGE_CACHE_MASK;
Fengguang Wud8983912007-07-19 01:48:06 -0700324 req_pages = (len + loff + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
Jens Axboe35f3d142010-05-20 10:43:18 +0200325 nr_pages = min(req_pages, pipe->buffers);
Jens Axboe5274f052006-03-30 15:15:30 +0200326
327 /*
Jens Axboeeb207962006-04-27 11:05:22 +0200328 * Lookup the (hopefully) full range of pages we need.
329 */
Jens Axboe35f3d142010-05-20 10:43:18 +0200330 spd.nr_pages = find_get_pages_contig(mapping, index, nr_pages, spd.pages);
Fengguang Wu431a48202007-07-19 01:48:05 -0700331 index += spd.nr_pages;
Jens Axboeeb207962006-04-27 11:05:22 +0200332
333 /*
334 * If find_get_pages_contig() returned fewer pages than we needed,
Fengguang Wu431a48202007-07-19 01:48:05 -0700335 * readahead/allocate the rest and fill in the holes.
Jens Axboeeb207962006-04-27 11:05:22 +0200336 */
Fengguang Wu431a48202007-07-19 01:48:05 -0700337 if (spd.nr_pages < nr_pages)
Rusty Russellcf914a72007-07-19 01:48:08 -0700338 page_cache_sync_readahead(mapping, &in->f_ra, in,
339 index, req_pages - spd.nr_pages);
Fengguang Wu431a48202007-07-19 01:48:05 -0700340
Jens Axboe932cc6d2007-06-21 13:10:21 +0200341 error = 0;
Jens Axboeeb207962006-04-27 11:05:22 +0200342 while (spd.nr_pages < nr_pages) {
343 /*
344 * Page could be there, find_get_pages_contig() breaks on
345 * the first hole.
346 */
347 page = find_get_page(mapping, index);
348 if (!page) {
Jens Axboee27dedd2006-05-01 19:59:54 +0200349 /*
Jens Axboeeb207962006-04-27 11:05:22 +0200350 * page didn't exist, allocate one.
351 */
352 page = page_cache_alloc_cold(mapping);
353 if (!page)
354 break;
355
356 error = add_to_page_cache_lru(page, mapping, index,
Nick Piggin0ae0b5d2010-05-25 10:25:26 +0200357 GFP_KERNEL);
Jens Axboeeb207962006-04-27 11:05:22 +0200358 if (unlikely(error)) {
359 page_cache_release(page);
Jens Axboea0548872006-05-03 10:58:22 +0200360 if (error == -EEXIST)
361 continue;
Jens Axboeeb207962006-04-27 11:05:22 +0200362 break;
363 }
364 /*
365 * add_to_page_cache() locks the page, unlock it
366 * to avoid convoluting the logic below even more.
367 */
368 unlock_page(page);
369 }
370
Jens Axboe35f3d142010-05-20 10:43:18 +0200371 spd.pages[spd.nr_pages++] = page;
Jens Axboeeb207962006-04-27 11:05:22 +0200372 index++;
373 }
374
375 /*
376 * Now loop over the map and see if we need to start IO on any
377 * pages, fill in the partial map, etc.
378 */
379 index = *ppos >> PAGE_CACHE_SHIFT;
380 nr_pages = spd.nr_pages;
381 spd.nr_pages = 0;
382 for (page_nr = 0; page_nr < nr_pages; page_nr++) {
Jens Axboe82aa5d62006-04-20 13:05:48 +0200383 unsigned int this_len;
384
385 if (!len)
386 break;
387
388 /*
389 * this_len is the max we'll use from this page
390 */
Andrew Mortonba5f5d92006-04-25 15:33:34 +0200391 this_len = min_t(unsigned long, len, PAGE_CACHE_SIZE - loff);
Jens Axboe35f3d142010-05-20 10:43:18 +0200392 page = spd.pages[page_nr];
Jens Axboe7480a902006-04-11 13:52:47 +0200393
Fengguang Wua08a1662007-07-19 01:48:03 -0700394 if (PageReadahead(page))
Rusty Russellcf914a72007-07-19 01:48:08 -0700395 page_cache_async_readahead(mapping, &in->f_ra, in,
Fengguang Wud8983912007-07-19 01:48:06 -0700396 page, index, req_pages - page_nr);
Fengguang Wua08a1662007-07-19 01:48:03 -0700397
Jens Axboe7480a902006-04-11 13:52:47 +0200398 /*
399 * If the page isn't uptodate, we may need to start io on it
400 */
401 if (!PageUptodate(page)) {
Miklos Szeredi69650312010-08-03 12:48:50 +0200402 lock_page(page);
Jens Axboe7480a902006-04-11 13:52:47 +0200403
404 /*
Miklos Szeredi32502b82008-07-04 09:35:17 +0200405 * Page was truncated, or invalidated by the
406 * filesystem. Redo the find/create, but this time the
407 * page is kept locked, so there's no chance of another
408 * race with truncate/invalidate.
Jens Axboe7480a902006-04-11 13:52:47 +0200409 */
410 if (!page->mapping) {
411 unlock_page(page);
Miklos Szeredi32502b82008-07-04 09:35:17 +0200412 page = find_or_create_page(mapping, index,
413 mapping_gfp_mask(mapping));
414
415 if (!page) {
416 error = -ENOMEM;
417 break;
418 }
Jens Axboe35f3d142010-05-20 10:43:18 +0200419 page_cache_release(spd.pages[page_nr]);
420 spd.pages[page_nr] = page;
Jens Axboe7480a902006-04-11 13:52:47 +0200421 }
422 /*
423 * page was already under io and is now done, great
424 */
425 if (PageUptodate(page)) {
426 unlock_page(page);
427 goto fill_it;
428 }
429
Jens Axboe7480a902006-04-11 13:52:47 +0200430 /*
431 * need to read in the page
432 */
433 error = mapping->a_ops->readpage(in, page);
Jens Axboe7480a902006-04-11 13:52:47 +0200434 if (unlikely(error)) {
Jens Axboeeb207962006-04-27 11:05:22 +0200435 /*
436 * We really should re-lookup the page here,
437 * but it complicates things a lot. Instead
438 * lets just do what we already stored, and
439 * we'll get it the next time we are called.
440 */
Jens Axboe7480a902006-04-11 13:52:47 +0200441 if (error == AOP_TRUNCATED_PAGE)
Jens Axboeeb207962006-04-27 11:05:22 +0200442 error = 0;
443
Jens Axboe7480a902006-04-11 13:52:47 +0200444 break;
445 }
Jens Axboe620a3242007-06-07 09:39:42 +0200446 }
447fill_it:
448 /*
449 * i_size must be checked after PageUptodate.
450 */
451 isize = i_size_read(mapping->host);
452 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
453 if (unlikely(!isize || index > end_index))
454 break;
455
456 /*
457 * if this is the last page, see if we need to shrink
458 * the length and stop
459 */
460 if (end_index == index) {
461 unsigned int plen;
Jens Axboe91ad66e2006-04-19 15:55:10 +0200462
463 /*
Jens Axboe620a3242007-06-07 09:39:42 +0200464 * max good bytes in this page
Jens Axboe91ad66e2006-04-19 15:55:10 +0200465 */
Jens Axboe620a3242007-06-07 09:39:42 +0200466 plen = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
467 if (plen <= loff)
Jens Axboe91ad66e2006-04-19 15:55:10 +0200468 break;
Jens Axboe91ad66e2006-04-19 15:55:10 +0200469
470 /*
Jens Axboe620a3242007-06-07 09:39:42 +0200471 * force quit after adding this page
Jens Axboe91ad66e2006-04-19 15:55:10 +0200472 */
Jens Axboe620a3242007-06-07 09:39:42 +0200473 this_len = min(this_len, plen - loff);
474 len = this_len;
Jens Axboe7480a902006-04-11 13:52:47 +0200475 }
Jens Axboe620a3242007-06-07 09:39:42 +0200476
Jens Axboe35f3d142010-05-20 10:43:18 +0200477 spd.partial[page_nr].offset = loff;
478 spd.partial[page_nr].len = this_len;
Jens Axboe82aa5d62006-04-20 13:05:48 +0200479 len -= this_len;
Jens Axboe91ad66e2006-04-19 15:55:10 +0200480 loff = 0;
Jens Axboeeb207962006-04-27 11:05:22 +0200481 spd.nr_pages++;
482 index++;
Jens Axboe5274f052006-03-30 15:15:30 +0200483 }
484
Jens Axboeeb207962006-04-27 11:05:22 +0200485 /*
Hugh Dickins475ecad2007-06-07 09:36:00 +0200486 * Release any pages at the end, if we quit early. 'page_nr' is how far
Jens Axboeeb207962006-04-27 11:05:22 +0200487 * we got, 'nr_pages' is how many pages are in the map.
488 */
489 while (page_nr < nr_pages)
Jens Axboe35f3d142010-05-20 10:43:18 +0200490 page_cache_release(spd.pages[page_nr++]);
Fengguang Wuf4e6b492007-10-16 01:24:33 -0700491 in->f_ra.prev_pos = (loff_t)index << PAGE_CACHE_SHIFT;
Jens Axboeeb207962006-04-27 11:05:22 +0200492
Jens Axboe912d35f2006-04-26 10:59:21 +0200493 if (spd.nr_pages)
Jens Axboe35f3d142010-05-20 10:43:18 +0200494 error = splice_to_pipe(pipe, &spd);
Jens Axboe5274f052006-03-30 15:15:30 +0200495
Jens Axboe35f3d142010-05-20 10:43:18 +0200496 splice_shrink_spd(pipe, &spd);
Jens Axboe7480a902006-04-11 13:52:47 +0200497 return error;
Jens Axboe5274f052006-03-30 15:15:30 +0200498}
499
Jens Axboe83f91352006-04-02 23:05:09 +0200500/**
501 * generic_file_splice_read - splice data from file to a pipe
502 * @in: file to splice from
Jens Axboe932cc6d2007-06-21 13:10:21 +0200503 * @ppos: position in @in
Jens Axboe83f91352006-04-02 23:05:09 +0200504 * @pipe: pipe to splice to
505 * @len: number of bytes to splice
506 * @flags: splice modifier flags
507 *
Jens Axboe932cc6d2007-06-21 13:10:21 +0200508 * Description:
509 * Will read pages from given file and fill them into a pipe. Can be
510 * used as long as the address_space operations for the source implements
511 * a readpage() hook.
512 *
Jens Axboe83f91352006-04-02 23:05:09 +0200513 */
Jens Axboecbb7e572006-04-11 14:57:50 +0200514ssize_t generic_file_splice_read(struct file *in, loff_t *ppos,
515 struct pipe_inode_info *pipe, size_t len,
516 unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +0200517{
Jens Axboed366d3982007-06-01 14:54:11 +0200518 loff_t isize, left;
Jens Axboe8191ecd2008-04-10 08:24:25 +0200519 int ret;
Jens Axboed366d3982007-06-01 14:54:11 +0200520
521 isize = i_size_read(in->f_mapping->host);
522 if (unlikely(*ppos >= isize))
523 return 0;
524
525 left = isize - *ppos;
526 if (unlikely(left < len))
527 len = left;
Jens Axboe5274f052006-03-30 15:15:30 +0200528
Jens Axboe8191ecd2008-04-10 08:24:25 +0200529 ret = __generic_file_splice_read(in, ppos, pipe, len, flags);
Miklos Szeredi723590e2009-08-15 08:43:22 +0200530 if (ret > 0) {
Jens Axboecbb7e572006-04-11 14:57:50 +0200531 *ppos += ret;
Miklos Szeredi723590e2009-08-15 08:43:22 +0200532 file_accessed(in);
533 }
Jens Axboe5274f052006-03-30 15:15:30 +0200534
535 return ret;
536}
Jens Axboe059a8f32006-04-02 23:06:05 +0200537EXPORT_SYMBOL(generic_file_splice_read);
538
Miklos Szeredi68181732009-05-07 15:37:36 +0200539static const struct pipe_buf_operations default_pipe_buf_ops = {
540 .can_merge = 0,
541 .map = generic_pipe_buf_map,
542 .unmap = generic_pipe_buf_unmap,
543 .confirm = generic_pipe_buf_confirm,
544 .release = generic_pipe_buf_release,
545 .steal = generic_pipe_buf_steal,
546 .get = generic_pipe_buf_get,
547};
548
549static ssize_t kernel_readv(struct file *file, const struct iovec *vec,
550 unsigned long vlen, loff_t offset)
551{
552 mm_segment_t old_fs;
553 loff_t pos = offset;
554 ssize_t res;
555
556 old_fs = get_fs();
557 set_fs(get_ds());
558 /* The cast to a user pointer is valid due to the set_fs() */
559 res = vfs_readv(file, (const struct iovec __user *)vec, vlen, &pos);
560 set_fs(old_fs);
561
562 return res;
563}
564
Miklos Szeredib2858d72009-05-19 11:37:46 +0200565static ssize_t kernel_write(struct file *file, const char *buf, size_t count,
566 loff_t pos)
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +0200567{
568 mm_segment_t old_fs;
569 ssize_t res;
570
571 old_fs = get_fs();
572 set_fs(get_ds());
573 /* The cast to a user pointer is valid due to the set_fs() */
Miklos Szeredib2858d72009-05-19 11:37:46 +0200574 res = vfs_write(file, (const char __user *)buf, count, &pos);
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +0200575 set_fs(old_fs);
576
577 return res;
578}
579
Miklos Szeredi68181732009-05-07 15:37:36 +0200580ssize_t default_file_splice_read(struct file *in, loff_t *ppos,
581 struct pipe_inode_info *pipe, size_t len,
582 unsigned int flags)
583{
584 unsigned int nr_pages;
585 unsigned int nr_freed;
586 size_t offset;
Jens Axboe35f3d142010-05-20 10:43:18 +0200587 struct page *pages[PIPE_DEF_BUFFERS];
588 struct partial_page partial[PIPE_DEF_BUFFERS];
589 struct iovec *vec, __vec[PIPE_DEF_BUFFERS];
Miklos Szeredi68181732009-05-07 15:37:36 +0200590 ssize_t res;
591 size_t this_len;
592 int error;
593 int i;
594 struct splice_pipe_desc spd = {
595 .pages = pages,
596 .partial = partial,
597 .flags = flags,
598 .ops = &default_pipe_buf_ops,
599 .spd_release = spd_release_page,
600 };
601
Jens Axboe35f3d142010-05-20 10:43:18 +0200602 if (splice_grow_spd(pipe, &spd))
603 return -ENOMEM;
604
605 res = -ENOMEM;
606 vec = __vec;
607 if (pipe->buffers > PIPE_DEF_BUFFERS) {
608 vec = kmalloc(pipe->buffers * sizeof(struct iovec), GFP_KERNEL);
609 if (!vec)
610 goto shrink_ret;
611 }
612
Miklos Szeredi68181732009-05-07 15:37:36 +0200613 offset = *ppos & ~PAGE_CACHE_MASK;
614 nr_pages = (len + offset + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
615
Jens Axboe35f3d142010-05-20 10:43:18 +0200616 for (i = 0; i < nr_pages && i < pipe->buffers && len; i++) {
Miklos Szeredi68181732009-05-07 15:37:36 +0200617 struct page *page;
618
Jens Axboe4f231222009-05-13 08:35:35 +0200619 page = alloc_page(GFP_USER);
Miklos Szeredi68181732009-05-07 15:37:36 +0200620 error = -ENOMEM;
621 if (!page)
622 goto err;
623
624 this_len = min_t(size_t, len, PAGE_CACHE_SIZE - offset);
Jens Axboe4f231222009-05-13 08:35:35 +0200625 vec[i].iov_base = (void __user *) page_address(page);
Miklos Szeredi68181732009-05-07 15:37:36 +0200626 vec[i].iov_len = this_len;
Jens Axboe35f3d142010-05-20 10:43:18 +0200627 spd.pages[i] = page;
Miklos Szeredi68181732009-05-07 15:37:36 +0200628 spd.nr_pages++;
629 len -= this_len;
630 offset = 0;
631 }
632
633 res = kernel_readv(in, vec, spd.nr_pages, *ppos);
Andrew Morton77f6bf52009-05-14 09:49:44 +0200634 if (res < 0) {
635 error = res;
Miklos Szeredi68181732009-05-07 15:37:36 +0200636 goto err;
Andrew Morton77f6bf52009-05-14 09:49:44 +0200637 }
Miklos Szeredi68181732009-05-07 15:37:36 +0200638
639 error = 0;
640 if (!res)
641 goto err;
642
643 nr_freed = 0;
644 for (i = 0; i < spd.nr_pages; i++) {
Miklos Szeredi68181732009-05-07 15:37:36 +0200645 this_len = min_t(size_t, vec[i].iov_len, res);
Jens Axboe35f3d142010-05-20 10:43:18 +0200646 spd.partial[i].offset = 0;
647 spd.partial[i].len = this_len;
Miklos Szeredi68181732009-05-07 15:37:36 +0200648 if (!this_len) {
Jens Axboe35f3d142010-05-20 10:43:18 +0200649 __free_page(spd.pages[i]);
650 spd.pages[i] = NULL;
Miklos Szeredi68181732009-05-07 15:37:36 +0200651 nr_freed++;
652 }
653 res -= this_len;
654 }
655 spd.nr_pages -= nr_freed;
656
657 res = splice_to_pipe(pipe, &spd);
658 if (res > 0)
659 *ppos += res;
660
Jens Axboe35f3d142010-05-20 10:43:18 +0200661shrink_ret:
662 if (vec != __vec)
663 kfree(vec);
664 splice_shrink_spd(pipe, &spd);
Miklos Szeredi68181732009-05-07 15:37:36 +0200665 return res;
666
667err:
Jens Axboe4f231222009-05-13 08:35:35 +0200668 for (i = 0; i < spd.nr_pages; i++)
Jens Axboe35f3d142010-05-20 10:43:18 +0200669 __free_page(spd.pages[i]);
Jens Axboe4f231222009-05-13 08:35:35 +0200670
Jens Axboe35f3d142010-05-20 10:43:18 +0200671 res = error;
672 goto shrink_ret;
Miklos Szeredi68181732009-05-07 15:37:36 +0200673}
674EXPORT_SYMBOL(default_file_splice_read);
675
Jens Axboe5274f052006-03-30 15:15:30 +0200676/*
Jens Axboe4f6f0bd2006-04-02 23:04:46 +0200677 * Send 'sd->len' bytes to socket from 'sd->file' at position 'sd->pos'
Jens Axboe016b6612006-04-25 15:42:00 +0200678 * using sendpage(). Return the number of bytes sent.
Jens Axboe5274f052006-03-30 15:15:30 +0200679 */
Jens Axboe76ad4d12006-05-03 10:41:33 +0200680static int pipe_to_sendpage(struct pipe_inode_info *pipe,
Jens Axboe5274f052006-03-30 15:15:30 +0200681 struct pipe_buffer *buf, struct splice_desc *sd)
682{
Jens Axboe6a14b902007-06-14 13:08:55 +0200683 struct file *file = sd->u.file;
Jens Axboe5274f052006-03-30 15:15:30 +0200684 loff_t pos = sd->pos;
Michał Mirosława8adbe32010-12-17 08:56:44 +0100685 int more;
Jens Axboe5274f052006-03-30 15:15:30 +0200686
Michał Mirosława8adbe32010-12-17 08:56:44 +0100687 if (!likely(file->f_op && file->f_op->sendpage))
688 return -EINVAL;
Jens Axboe5274f052006-03-30 15:15:30 +0200689
Michał Mirosława8adbe32010-12-17 08:56:44 +0100690 more = (sd->flags & SPLICE_F_MORE) || sd->len < sd->total_len;
691 return file->f_op->sendpage(file, buf->page, buf->offset,
692 sd->len, &pos, more);
Jens Axboe5274f052006-03-30 15:15:30 +0200693}
694
695/*
696 * This is a little more tricky than the file -> pipe splicing. There are
697 * basically three cases:
698 *
699 * - Destination page already exists in the address space and there
700 * are users of it. For that case we have no other option that
701 * copying the data. Tough luck.
702 * - Destination page already exists in the address space, but there
703 * are no users of it. Make sure it's uptodate, then drop it. Fall
704 * through to last case.
705 * - Destination page does not exist, we can add the pipe page to
706 * the page cache and avoid the copy.
707 *
Jens Axboe83f91352006-04-02 23:05:09 +0200708 * If asked to move pages to the output file (SPLICE_F_MOVE is set in
709 * sd->flags), we attempt to migrate pages from the pipe to the output
710 * file address space page cache. This is possible if no one else has
711 * the pipe page referenced outside of the pipe and page cache. If
712 * SPLICE_F_MOVE isn't set, or we cannot move the page, we simply create
713 * a new page in the output file page cache and fill/dirty that.
Jens Axboe5274f052006-03-30 15:15:30 +0200714 */
Miklos Szeredi328eaab2009-04-14 19:48:39 +0200715int pipe_to_file(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
716 struct splice_desc *sd)
Jens Axboe5274f052006-03-30 15:15:30 +0200717{
Jens Axboe6a14b902007-06-14 13:08:55 +0200718 struct file *file = sd->u.file;
Jens Axboe5274f052006-03-30 15:15:30 +0200719 struct address_space *mapping = file->f_mapping;
Jens Axboe016b6612006-04-25 15:42:00 +0200720 unsigned int offset, this_len;
Jens Axboe5274f052006-03-30 15:15:30 +0200721 struct page *page;
Nick Pigginafddba42007-10-16 01:25:01 -0700722 void *fsdata;
Jens Axboe3e7ee3e2006-04-02 23:11:04 +0200723 int ret;
Jens Axboe5274f052006-03-30 15:15:30 +0200724
Jens Axboe5274f052006-03-30 15:15:30 +0200725 offset = sd->pos & ~PAGE_CACHE_MASK;
726
Jens Axboe016b6612006-04-25 15:42:00 +0200727 this_len = sd->len;
728 if (this_len + offset > PAGE_CACHE_SIZE)
729 this_len = PAGE_CACHE_SIZE - offset;
730
Nick Pigginafddba42007-10-16 01:25:01 -0700731 ret = pagecache_write_begin(file, mapping, sd->pos, this_len,
732 AOP_FLAG_UNINTERRUPTIBLE, &page, &fsdata);
733 if (unlikely(ret))
734 goto out;
Jens Axboe5274f052006-03-30 15:15:30 +0200735
Jens Axboe0568b402006-05-01 19:50:48 +0200736 if (buf->page != page) {
Jens Axboef84d7512006-05-01 19:59:03 +0200737 /*
738 * Careful, ->map() uses KM_USER0!
739 */
Jens Axboe76ad4d12006-05-03 10:41:33 +0200740 char *src = buf->ops->map(pipe, buf, 1);
Jens Axboef84d7512006-05-01 19:59:03 +0200741 char *dst = kmap_atomic(page, KM_USER1);
Jens Axboe5abc97a2006-03-30 15:16:46 +0200742
Jens Axboe016b6612006-04-25 15:42:00 +0200743 memcpy(dst + offset, src + buf->offset, this_len);
Jens Axboe5abc97a2006-03-30 15:16:46 +0200744 flush_dcache_page(page);
Jens Axboef84d7512006-05-01 19:59:03 +0200745 kunmap_atomic(dst, KM_USER1);
Jens Axboe76ad4d12006-05-03 10:41:33 +0200746 buf->ops->unmap(pipe, buf, src);
Jens Axboe5abc97a2006-03-30 15:16:46 +0200747 }
Nick Pigginafddba42007-10-16 01:25:01 -0700748 ret = pagecache_write_end(file, mapping, sd->pos, this_len, this_len,
749 page, fsdata);
Jens Axboe5274f052006-03-30 15:15:30 +0200750out:
Jens Axboe5274f052006-03-30 15:15:30 +0200751 return ret;
752}
Miklos Szeredi328eaab2009-04-14 19:48:39 +0200753EXPORT_SYMBOL(pipe_to_file);
Jens Axboe5274f052006-03-30 15:15:30 +0200754
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200755static void wakeup_pipe_writers(struct pipe_inode_info *pipe)
756{
757 smp_mb();
758 if (waitqueue_active(&pipe->wait))
759 wake_up_interruptible(&pipe->wait);
760 kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
761}
762
763/**
764 * splice_from_pipe_feed - feed available data from a pipe to a file
765 * @pipe: pipe to splice from
766 * @sd: information to @actor
767 * @actor: handler that splices the data
768 *
769 * Description:
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200770 * This function loops over the pipe and calls @actor to do the
771 * actual moving of a single struct pipe_buffer to the desired
772 * destination. It returns when there's no more buffers left in
773 * the pipe or if the requested number of bytes (@sd->total_len)
774 * have been copied. It returns a positive number (one) if the
775 * pipe needs to be filled with more data, zero if the required
776 * number of bytes have been copied and -errno on error.
777 *
778 * This, together with splice_from_pipe_{begin,end,next}, may be
779 * used to implement the functionality of __splice_from_pipe() when
780 * locking is required around copying the pipe buffers to the
781 * destination.
782 */
783int splice_from_pipe_feed(struct pipe_inode_info *pipe, struct splice_desc *sd,
784 splice_actor *actor)
785{
786 int ret;
787
788 while (pipe->nrbufs) {
789 struct pipe_buffer *buf = pipe->bufs + pipe->curbuf;
790 const struct pipe_buf_operations *ops = buf->ops;
791
792 sd->len = buf->len;
793 if (sd->len > sd->total_len)
794 sd->len = sd->total_len;
795
Michał Mirosława8adbe32010-12-17 08:56:44 +0100796 ret = buf->ops->confirm(pipe, buf);
797 if (unlikely(ret)) {
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200798 if (ret == -ENODATA)
799 ret = 0;
800 return ret;
801 }
Michał Mirosława8adbe32010-12-17 08:56:44 +0100802
803 ret = actor(pipe, buf, sd);
804 if (ret <= 0)
805 return ret;
806
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200807 buf->offset += ret;
808 buf->len -= ret;
809
810 sd->num_spliced += ret;
811 sd->len -= ret;
812 sd->pos += ret;
813 sd->total_len -= ret;
814
815 if (!buf->len) {
816 buf->ops = NULL;
817 ops->release(pipe, buf);
Jens Axboe35f3d142010-05-20 10:43:18 +0200818 pipe->curbuf = (pipe->curbuf + 1) & (pipe->buffers - 1);
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200819 pipe->nrbufs--;
820 if (pipe->inode)
821 sd->need_wakeup = true;
822 }
823
824 if (!sd->total_len)
825 return 0;
826 }
827
828 return 1;
829}
830EXPORT_SYMBOL(splice_from_pipe_feed);
831
832/**
833 * splice_from_pipe_next - wait for some data to splice from
834 * @pipe: pipe to splice from
835 * @sd: information about the splice operation
836 *
837 * Description:
838 * This function will wait for some data and return a positive
839 * value (one) if pipe buffers are available. It will return zero
840 * or -errno if no more data needs to be spliced.
841 */
842int splice_from_pipe_next(struct pipe_inode_info *pipe, struct splice_desc *sd)
843{
844 while (!pipe->nrbufs) {
845 if (!pipe->writers)
846 return 0;
847
848 if (!pipe->waiting_writers && sd->num_spliced)
849 return 0;
850
851 if (sd->flags & SPLICE_F_NONBLOCK)
852 return -EAGAIN;
853
854 if (signal_pending(current))
855 return -ERESTARTSYS;
856
857 if (sd->need_wakeup) {
858 wakeup_pipe_writers(pipe);
859 sd->need_wakeup = false;
860 }
861
862 pipe_wait(pipe);
863 }
864
865 return 1;
866}
867EXPORT_SYMBOL(splice_from_pipe_next);
868
869/**
870 * splice_from_pipe_begin - start splicing from pipe
Randy Dunlapb80901b2009-04-16 19:09:55 -0700871 * @sd: information about the splice operation
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200872 *
873 * Description:
874 * This function should be called before a loop containing
875 * splice_from_pipe_next() and splice_from_pipe_feed() to
876 * initialize the necessary fields of @sd.
877 */
878void splice_from_pipe_begin(struct splice_desc *sd)
879{
880 sd->num_spliced = 0;
881 sd->need_wakeup = false;
882}
883EXPORT_SYMBOL(splice_from_pipe_begin);
884
885/**
886 * splice_from_pipe_end - finish splicing from pipe
887 * @pipe: pipe to splice from
888 * @sd: information about the splice operation
889 *
890 * Description:
891 * This function will wake up pipe writers if necessary. It should
892 * be called after a loop containing splice_from_pipe_next() and
893 * splice_from_pipe_feed().
894 */
895void splice_from_pipe_end(struct pipe_inode_info *pipe, struct splice_desc *sd)
896{
897 if (sd->need_wakeup)
898 wakeup_pipe_writers(pipe);
899}
900EXPORT_SYMBOL(splice_from_pipe_end);
901
Jens Axboe932cc6d2007-06-21 13:10:21 +0200902/**
903 * __splice_from_pipe - splice data from a pipe to given actor
904 * @pipe: pipe to splice from
905 * @sd: information to @actor
906 * @actor: handler that splices the data
907 *
908 * Description:
909 * This function does little more than loop over the pipe and call
910 * @actor to do the actual moving of a single struct pipe_buffer to
911 * the desired destination. See pipe_to_file, pipe_to_sendpage, or
912 * pipe_to_user.
913 *
Jens Axboe83f91352006-04-02 23:05:09 +0200914 */
Jens Axboec66ab6f2007-06-12 21:17:17 +0200915ssize_t __splice_from_pipe(struct pipe_inode_info *pipe, struct splice_desc *sd,
916 splice_actor *actor)
Jens Axboe5274f052006-03-30 15:15:30 +0200917{
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200918 int ret;
Jens Axboe5274f052006-03-30 15:15:30 +0200919
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200920 splice_from_pipe_begin(sd);
921 do {
922 ret = splice_from_pipe_next(pipe, sd);
923 if (ret > 0)
924 ret = splice_from_pipe_feed(pipe, sd, actor);
925 } while (ret > 0);
926 splice_from_pipe_end(pipe, sd);
Jens Axboe5274f052006-03-30 15:15:30 +0200927
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200928 return sd->num_spliced ? sd->num_spliced : ret;
Jens Axboe5274f052006-03-30 15:15:30 +0200929}
Mark Fasheh40bee44e2007-03-21 13:11:02 +0100930EXPORT_SYMBOL(__splice_from_pipe);
Jens Axboe5274f052006-03-30 15:15:30 +0200931
Jens Axboe932cc6d2007-06-21 13:10:21 +0200932/**
933 * splice_from_pipe - splice data from a pipe to a file
934 * @pipe: pipe to splice from
935 * @out: file to splice to
936 * @ppos: position in @out
937 * @len: how many bytes to splice
938 * @flags: splice modifier flags
939 * @actor: handler that splices the data
940 *
941 * Description:
Miklos Szeredi29339702009-04-14 19:48:37 +0200942 * See __splice_from_pipe. This function locks the pipe inode,
Jens Axboe932cc6d2007-06-21 13:10:21 +0200943 * otherwise it's identical to __splice_from_pipe().
944 *
945 */
Mark Fasheh6da61802006-10-17 18:43:07 +0200946ssize_t splice_from_pipe(struct pipe_inode_info *pipe, struct file *out,
947 loff_t *ppos, size_t len, unsigned int flags,
948 splice_actor *actor)
949{
950 ssize_t ret;
Jens Axboec66ab6f2007-06-12 21:17:17 +0200951 struct splice_desc sd = {
952 .total_len = len,
953 .flags = flags,
954 .pos = *ppos,
Jens Axboe6a14b902007-06-14 13:08:55 +0200955 .u.file = out,
Jens Axboec66ab6f2007-06-12 21:17:17 +0200956 };
Mark Fasheh6da61802006-10-17 18:43:07 +0200957
Miklos Szeredi61e0d472009-04-14 19:48:41 +0200958 pipe_lock(pipe);
Jens Axboec66ab6f2007-06-12 21:17:17 +0200959 ret = __splice_from_pipe(pipe, &sd, actor);
Miklos Szeredi61e0d472009-04-14 19:48:41 +0200960 pipe_unlock(pipe);
Mark Fasheh6da61802006-10-17 18:43:07 +0200961
962 return ret;
963}
964
965/**
Jens Axboe83f91352006-04-02 23:05:09 +0200966 * generic_file_splice_write - splice data from a pipe to a file
Ingo Molnar3a326a22006-04-10 15:18:35 +0200967 * @pipe: pipe info
Jens Axboe83f91352006-04-02 23:05:09 +0200968 * @out: file to write to
Jens Axboe932cc6d2007-06-21 13:10:21 +0200969 * @ppos: position in @out
Jens Axboe83f91352006-04-02 23:05:09 +0200970 * @len: number of bytes to splice
971 * @flags: splice modifier flags
972 *
Jens Axboe932cc6d2007-06-21 13:10:21 +0200973 * Description:
974 * Will either move or copy pages (determined by @flags options) from
975 * the given pipe inode to the given file.
Jens Axboe83f91352006-04-02 23:05:09 +0200976 *
977 */
Ingo Molnar3a326a22006-04-10 15:18:35 +0200978ssize_t
979generic_file_splice_write(struct pipe_inode_info *pipe, struct file *out,
Jens Axboecbb7e572006-04-11 14:57:50 +0200980 loff_t *ppos, size_t len, unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +0200981{
Jens Axboe4f6f0bd2006-04-02 23:04:46 +0200982 struct address_space *mapping = out->f_mapping;
Jens Axboe8c34e2d2006-10-17 19:43:22 +0200983 struct inode *inode = mapping->host;
Miklos Szeredi7f3d4ee2008-05-07 09:22:39 +0200984 struct splice_desc sd = {
985 .total_len = len,
986 .flags = flags,
987 .pos = *ppos,
988 .u.file = out,
989 };
Ingo Molnar3a326a22006-04-10 15:18:35 +0200990 ssize_t ret;
Jens Axboe8c34e2d2006-10-17 19:43:22 +0200991
Miklos Szeredi61e0d472009-04-14 19:48:41 +0200992 pipe_lock(pipe);
Miklos Szeredieb443e52009-04-14 19:48:38 +0200993
994 splice_from_pipe_begin(&sd);
995 do {
996 ret = splice_from_pipe_next(pipe, &sd);
997 if (ret <= 0)
998 break;
999
1000 mutex_lock_nested(&inode->i_mutex, I_MUTEX_CHILD);
1001 ret = file_remove_suid(out);
Miklos Szeredi723590e2009-08-15 08:43:22 +02001002 if (!ret) {
1003 file_update_time(out);
Miklos Szeredieb443e52009-04-14 19:48:38 +02001004 ret = splice_from_pipe_feed(pipe, &sd, pipe_to_file);
Miklos Szeredi723590e2009-08-15 08:43:22 +02001005 }
Miklos Szeredieb443e52009-04-14 19:48:38 +02001006 mutex_unlock(&inode->i_mutex);
1007 } while (ret > 0);
1008 splice_from_pipe_end(pipe, &sd);
1009
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001010 pipe_unlock(pipe);
Miklos Szeredieb443e52009-04-14 19:48:38 +02001011
1012 if (sd.num_spliced)
1013 ret = sd.num_spliced;
1014
Jens Axboea4514eb2006-04-19 15:57:05 +02001015 if (ret > 0) {
Jens Axboe17ee4f42007-06-15 13:10:37 +02001016 unsigned long nr_pages;
Jan Kara148f9482009-08-17 19:52:36 +02001017 int err;
Jens Axboe17ee4f42007-06-15 13:10:37 +02001018
Jens Axboe17ee4f42007-06-15 13:10:37 +02001019 nr_pages = (ret + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
Jens Axboe4f6f0bd2006-04-02 23:04:46 +02001020
Jan Kara148f9482009-08-17 19:52:36 +02001021 err = generic_write_sync(out, *ppos, ret);
1022 if (err)
1023 ret = err;
1024 else
1025 *ppos += ret;
Jens Axboe17ee4f42007-06-15 13:10:37 +02001026 balance_dirty_pages_ratelimited_nr(mapping, nr_pages);
Jens Axboe4f6f0bd2006-04-02 23:04:46 +02001027 }
1028
1029 return ret;
Jens Axboe5274f052006-03-30 15:15:30 +02001030}
1031
Jens Axboe059a8f32006-04-02 23:06:05 +02001032EXPORT_SYMBOL(generic_file_splice_write);
1033
Miklos Szeredib2858d72009-05-19 11:37:46 +02001034static int write_pipe_buf(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
1035 struct splice_desc *sd)
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +02001036{
Miklos Szeredib2858d72009-05-19 11:37:46 +02001037 int ret;
1038 void *data;
1039
Miklos Szeredib2858d72009-05-19 11:37:46 +02001040 data = buf->ops->map(pipe, buf, 0);
1041 ret = kernel_write(sd->u.file, data + buf->offset, sd->len, sd->pos);
1042 buf->ops->unmap(pipe, buf, data);
1043
1044 return ret;
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +02001045}
1046
1047static ssize_t default_file_splice_write(struct pipe_inode_info *pipe,
1048 struct file *out, loff_t *ppos,
1049 size_t len, unsigned int flags)
1050{
Miklos Szeredib2858d72009-05-19 11:37:46 +02001051 ssize_t ret;
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +02001052
Miklos Szeredib2858d72009-05-19 11:37:46 +02001053 ret = splice_from_pipe(pipe, out, ppos, len, flags, write_pipe_buf);
1054 if (ret > 0)
1055 *ppos += ret;
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +02001056
Miklos Szeredib2858d72009-05-19 11:37:46 +02001057 return ret;
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +02001058}
1059
Jens Axboe83f91352006-04-02 23:05:09 +02001060/**
1061 * generic_splice_sendpage - splice data from a pipe to a socket
Jens Axboe932cc6d2007-06-21 13:10:21 +02001062 * @pipe: pipe to splice from
Jens Axboe83f91352006-04-02 23:05:09 +02001063 * @out: socket to write to
Jens Axboe932cc6d2007-06-21 13:10:21 +02001064 * @ppos: position in @out
Jens Axboe83f91352006-04-02 23:05:09 +02001065 * @len: number of bytes to splice
1066 * @flags: splice modifier flags
1067 *
Jens Axboe932cc6d2007-06-21 13:10:21 +02001068 * Description:
1069 * Will send @len bytes from the pipe to a network socket. No data copying
1070 * is involved.
Jens Axboe83f91352006-04-02 23:05:09 +02001071 *
1072 */
Ingo Molnar3a326a22006-04-10 15:18:35 +02001073ssize_t generic_splice_sendpage(struct pipe_inode_info *pipe, struct file *out,
Jens Axboecbb7e572006-04-11 14:57:50 +02001074 loff_t *ppos, size_t len, unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +02001075{
Jens Axboe00522fb2006-04-26 14:39:29 +02001076 return splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_sendpage);
Jens Axboe5274f052006-03-30 15:15:30 +02001077}
1078
Jens Axboe059a8f32006-04-02 23:06:05 +02001079EXPORT_SYMBOL(generic_splice_sendpage);
Jeff Garzika0f06782006-03-30 23:06:13 -05001080
Jens Axboe83f91352006-04-02 23:05:09 +02001081/*
1082 * Attempt to initiate a splice from pipe to file.
1083 */
Ingo Molnar3a326a22006-04-10 15:18:35 +02001084static long do_splice_from(struct pipe_inode_info *pipe, struct file *out,
Jens Axboecbb7e572006-04-11 14:57:50 +02001085 loff_t *ppos, size_t len, unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +02001086{
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +02001087 ssize_t (*splice_write)(struct pipe_inode_info *, struct file *,
1088 loff_t *, size_t, unsigned int);
Jens Axboe5274f052006-03-30 15:15:30 +02001089 int ret;
1090
Jens Axboe49570e92006-04-11 13:56:09 +02001091 if (unlikely(!(out->f_mode & FMODE_WRITE)))
Jens Axboe5274f052006-03-30 15:15:30 +02001092 return -EBADF;
1093
Linus Torvaldsefc968d2008-10-09 14:04:54 -07001094 if (unlikely(out->f_flags & O_APPEND))
1095 return -EINVAL;
1096
Jens Axboecbb7e572006-04-11 14:57:50 +02001097 ret = rw_verify_area(WRITE, out, ppos, len);
Jens Axboe5274f052006-03-30 15:15:30 +02001098 if (unlikely(ret < 0))
1099 return ret;
1100
Changli Gaocc56f7d2009-11-04 09:09:52 +01001101 if (out->f_op && out->f_op->splice_write)
1102 splice_write = out->f_op->splice_write;
1103 else
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +02001104 splice_write = default_file_splice_write;
1105
1106 return splice_write(pipe, out, ppos, len, flags);
Jens Axboe5274f052006-03-30 15:15:30 +02001107}
1108
Jens Axboe83f91352006-04-02 23:05:09 +02001109/*
1110 * Attempt to initiate a splice from a file to a pipe.
1111 */
Jens Axboecbb7e572006-04-11 14:57:50 +02001112static long do_splice_to(struct file *in, loff_t *ppos,
1113 struct pipe_inode_info *pipe, size_t len,
1114 unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +02001115{
Miklos Szeredi68181732009-05-07 15:37:36 +02001116 ssize_t (*splice_read)(struct file *, loff_t *,
1117 struct pipe_inode_info *, size_t, unsigned int);
Jens Axboe5274f052006-03-30 15:15:30 +02001118 int ret;
1119
Jens Axboe49570e92006-04-11 13:56:09 +02001120 if (unlikely(!(in->f_mode & FMODE_READ)))
Jens Axboe5274f052006-03-30 15:15:30 +02001121 return -EBADF;
1122
Jens Axboecbb7e572006-04-11 14:57:50 +02001123 ret = rw_verify_area(READ, in, ppos, len);
Jens Axboe5274f052006-03-30 15:15:30 +02001124 if (unlikely(ret < 0))
1125 return ret;
1126
Changli Gaocc56f7d2009-11-04 09:09:52 +01001127 if (in->f_op && in->f_op->splice_read)
1128 splice_read = in->f_op->splice_read;
1129 else
Miklos Szeredi68181732009-05-07 15:37:36 +02001130 splice_read = default_file_splice_read;
1131
1132 return splice_read(in, ppos, pipe, len, flags);
Jens Axboe5274f052006-03-30 15:15:30 +02001133}
1134
Jens Axboe932cc6d2007-06-21 13:10:21 +02001135/**
1136 * splice_direct_to_actor - splices data directly between two non-pipes
1137 * @in: file to splice from
1138 * @sd: actor information on where to splice to
1139 * @actor: handles the data splicing
1140 *
1141 * Description:
1142 * This is a special case helper to splice directly between two
1143 * points, without requiring an explicit pipe. Internally an allocated
Randy Dunlap79685b82007-07-27 08:08:51 +02001144 * pipe is cached in the process, and reused during the lifetime of
Jens Axboe932cc6d2007-06-21 13:10:21 +02001145 * that process.
1146 *
Jens Axboec66ab6f2007-06-12 21:17:17 +02001147 */
1148ssize_t splice_direct_to_actor(struct file *in, struct splice_desc *sd,
1149 splice_direct_actor *actor)
Jens Axboeb92ce552006-04-11 13:52:07 +02001150{
1151 struct pipe_inode_info *pipe;
1152 long ret, bytes;
1153 umode_t i_mode;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001154 size_t len;
1155 int i, flags;
Jens Axboeb92ce552006-04-11 13:52:07 +02001156
1157 /*
1158 * We require the input being a regular file, as we don't want to
1159 * randomly drop data for eg socket -> socket splicing. Use the
1160 * piped splicing for that!
1161 */
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001162 i_mode = in->f_path.dentry->d_inode->i_mode;
Jens Axboeb92ce552006-04-11 13:52:07 +02001163 if (unlikely(!S_ISREG(i_mode) && !S_ISBLK(i_mode)))
1164 return -EINVAL;
1165
1166 /*
1167 * neither in nor out is a pipe, setup an internal pipe attached to
1168 * 'out' and transfer the wanted data from 'in' to 'out' through that
1169 */
1170 pipe = current->splice_pipe;
Jens Axboe49570e92006-04-11 13:56:09 +02001171 if (unlikely(!pipe)) {
Jens Axboeb92ce552006-04-11 13:52:07 +02001172 pipe = alloc_pipe_info(NULL);
1173 if (!pipe)
1174 return -ENOMEM;
1175
1176 /*
1177 * We don't have an immediate reader, but we'll read the stuff
Jens Axboe00522fb2006-04-26 14:39:29 +02001178 * out of the pipe right after the splice_to_pipe(). So set
Jens Axboeb92ce552006-04-11 13:52:07 +02001179 * PIPE_READERS appropriately.
1180 */
1181 pipe->readers = 1;
1182
1183 current->splice_pipe = pipe;
1184 }
1185
1186 /*
Ingo Molnar73d62d82006-04-11 13:57:21 +02001187 * Do the splice.
Jens Axboeb92ce552006-04-11 13:52:07 +02001188 */
1189 ret = 0;
1190 bytes = 0;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001191 len = sd->total_len;
1192 flags = sd->flags;
1193
1194 /*
1195 * Don't block on output, we have to drain the direct pipe.
1196 */
1197 sd->flags &= ~SPLICE_F_NONBLOCK;
Jens Axboeb92ce552006-04-11 13:52:07 +02001198
1199 while (len) {
Jens Axboe51a92c02007-07-13 14:11:43 +02001200 size_t read_len;
Tom Zanussia82c53a2008-05-09 13:28:36 +02001201 loff_t pos = sd->pos, prev_pos = pos;
Jens Axboeb92ce552006-04-11 13:52:07 +02001202
Jens Axboebcd4f3a2007-07-16 14:41:49 +02001203 ret = do_splice_to(in, &pos, pipe, len, flags);
Jens Axboe51a92c02007-07-13 14:11:43 +02001204 if (unlikely(ret <= 0))
Jens Axboeb92ce552006-04-11 13:52:07 +02001205 goto out_release;
1206
1207 read_len = ret;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001208 sd->total_len = read_len;
Jens Axboeb92ce552006-04-11 13:52:07 +02001209
1210 /*
1211 * NOTE: nonblocking mode only applies to the input. We
1212 * must not do the output in nonblocking mode as then we
1213 * could get stuck data in the internal pipe:
1214 */
Jens Axboec66ab6f2007-06-12 21:17:17 +02001215 ret = actor(pipe, sd);
Tom Zanussia82c53a2008-05-09 13:28:36 +02001216 if (unlikely(ret <= 0)) {
1217 sd->pos = prev_pos;
Jens Axboeb92ce552006-04-11 13:52:07 +02001218 goto out_release;
Tom Zanussia82c53a2008-05-09 13:28:36 +02001219 }
Jens Axboeb92ce552006-04-11 13:52:07 +02001220
1221 bytes += ret;
1222 len -= ret;
Jens Axboebcd4f3a2007-07-16 14:41:49 +02001223 sd->pos = pos;
Jens Axboeb92ce552006-04-11 13:52:07 +02001224
Tom Zanussia82c53a2008-05-09 13:28:36 +02001225 if (ret < read_len) {
1226 sd->pos = prev_pos + ret;
Jens Axboe51a92c02007-07-13 14:11:43 +02001227 goto out_release;
Tom Zanussia82c53a2008-05-09 13:28:36 +02001228 }
Jens Axboeb92ce552006-04-11 13:52:07 +02001229 }
1230
Jens Axboe9e971982008-01-29 21:05:57 +01001231done:
Jens Axboeb92ce552006-04-11 13:52:07 +02001232 pipe->nrbufs = pipe->curbuf = 0;
Jens Axboe80848702008-01-30 12:24:48 +01001233 file_accessed(in);
Jens Axboeb92ce552006-04-11 13:52:07 +02001234 return bytes;
1235
1236out_release:
1237 /*
1238 * If we did an incomplete transfer we must release
1239 * the pipe buffers in question:
1240 */
Jens Axboe35f3d142010-05-20 10:43:18 +02001241 for (i = 0; i < pipe->buffers; i++) {
Jens Axboeb92ce552006-04-11 13:52:07 +02001242 struct pipe_buffer *buf = pipe->bufs + i;
1243
1244 if (buf->ops) {
1245 buf->ops->release(pipe, buf);
1246 buf->ops = NULL;
1247 }
1248 }
Jens Axboeb92ce552006-04-11 13:52:07 +02001249
Jens Axboe9e971982008-01-29 21:05:57 +01001250 if (!bytes)
1251 bytes = ret;
Jens Axboeb92ce552006-04-11 13:52:07 +02001252
Jens Axboe9e971982008-01-29 21:05:57 +01001253 goto done;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001254}
1255EXPORT_SYMBOL(splice_direct_to_actor);
1256
1257static int direct_splice_actor(struct pipe_inode_info *pipe,
1258 struct splice_desc *sd)
1259{
Jens Axboe6a14b902007-06-14 13:08:55 +02001260 struct file *file = sd->u.file;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001261
Changli Gao2cb4b052010-06-29 13:09:18 +02001262 return do_splice_from(pipe, file, &file->f_pos, sd->total_len,
1263 sd->flags);
Jens Axboec66ab6f2007-06-12 21:17:17 +02001264}
1265
Jens Axboe932cc6d2007-06-21 13:10:21 +02001266/**
1267 * do_splice_direct - splices data directly between two files
1268 * @in: file to splice from
1269 * @ppos: input file offset
1270 * @out: file to splice to
1271 * @len: number of bytes to splice
1272 * @flags: splice modifier flags
1273 *
1274 * Description:
1275 * For use by do_sendfile(). splice can easily emulate sendfile, but
1276 * doing it in the application would incur an extra system call
1277 * (splice in + splice out, as compared to just sendfile()). So this helper
1278 * can splice directly through a process-private pipe.
1279 *
1280 */
Jens Axboec66ab6f2007-06-12 21:17:17 +02001281long do_splice_direct(struct file *in, loff_t *ppos, struct file *out,
1282 size_t len, unsigned int flags)
1283{
1284 struct splice_desc sd = {
1285 .len = len,
1286 .total_len = len,
1287 .flags = flags,
1288 .pos = *ppos,
Jens Axboe6a14b902007-06-14 13:08:55 +02001289 .u.file = out,
Jens Axboec66ab6f2007-06-12 21:17:17 +02001290 };
Jens Axboe51a92c02007-07-13 14:11:43 +02001291 long ret;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001292
1293 ret = splice_direct_to_actor(in, &sd, direct_splice_actor);
Jens Axboe51a92c02007-07-13 14:11:43 +02001294 if (ret > 0)
Tom Zanussia82c53a2008-05-09 13:28:36 +02001295 *ppos = sd.pos;
Jens Axboe51a92c02007-07-13 14:11:43 +02001296
Jens Axboec66ab6f2007-06-12 21:17:17 +02001297 return ret;
Jens Axboeb92ce552006-04-11 13:52:07 +02001298}
1299
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001300static int splice_pipe_to_pipe(struct pipe_inode_info *ipipe,
1301 struct pipe_inode_info *opipe,
1302 size_t len, unsigned int flags);
Jens Axboeddac0d32006-11-04 12:49:32 +01001303
1304/*
Jens Axboe83f91352006-04-02 23:05:09 +02001305 * Determine where to splice to/from.
1306 */
Ingo Molnar529565d2006-04-10 15:18:58 +02001307static long do_splice(struct file *in, loff_t __user *off_in,
1308 struct file *out, loff_t __user *off_out,
1309 size_t len, unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +02001310{
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001311 struct pipe_inode_info *ipipe;
1312 struct pipe_inode_info *opipe;
Jens Axboecbb7e572006-04-11 14:57:50 +02001313 loff_t offset, *off;
Jens Axboea4514eb2006-04-19 15:57:05 +02001314 long ret;
Jens Axboe5274f052006-03-30 15:15:30 +02001315
Linus Torvalds71993e62010-11-28 13:56:09 -08001316 ipipe = get_pipe_info(in);
1317 opipe = get_pipe_info(out);
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001318
1319 if (ipipe && opipe) {
1320 if (off_in || off_out)
1321 return -ESPIPE;
1322
1323 if (!(in->f_mode & FMODE_READ))
1324 return -EBADF;
1325
1326 if (!(out->f_mode & FMODE_WRITE))
1327 return -EBADF;
1328
1329 /* Splicing to self would be fun, but... */
1330 if (ipipe == opipe)
1331 return -EINVAL;
1332
1333 return splice_pipe_to_pipe(ipipe, opipe, len, flags);
1334 }
1335
1336 if (ipipe) {
Ingo Molnar529565d2006-04-10 15:18:58 +02001337 if (off_in)
1338 return -ESPIPE;
Jens Axboeb92ce552006-04-11 13:52:07 +02001339 if (off_out) {
Changli Gao19c9a492010-06-29 13:10:36 +02001340 if (!(out->f_mode & FMODE_PWRITE))
Jens Axboeb92ce552006-04-11 13:52:07 +02001341 return -EINVAL;
Jens Axboecbb7e572006-04-11 14:57:50 +02001342 if (copy_from_user(&offset, off_out, sizeof(loff_t)))
Jens Axboeb92ce552006-04-11 13:52:07 +02001343 return -EFAULT;
Jens Axboecbb7e572006-04-11 14:57:50 +02001344 off = &offset;
1345 } else
1346 off = &out->f_pos;
Ingo Molnar529565d2006-04-10 15:18:58 +02001347
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001348 ret = do_splice_from(ipipe, out, off, len, flags);
Jens Axboea4514eb2006-04-19 15:57:05 +02001349
1350 if (off_out && copy_to_user(off_out, off, sizeof(loff_t)))
1351 ret = -EFAULT;
1352
1353 return ret;
Ingo Molnar529565d2006-04-10 15:18:58 +02001354 }
Jens Axboe5274f052006-03-30 15:15:30 +02001355
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001356 if (opipe) {
Ingo Molnar529565d2006-04-10 15:18:58 +02001357 if (off_out)
1358 return -ESPIPE;
Jens Axboeb92ce552006-04-11 13:52:07 +02001359 if (off_in) {
Changli Gao19c9a492010-06-29 13:10:36 +02001360 if (!(in->f_mode & FMODE_PREAD))
Jens Axboeb92ce552006-04-11 13:52:07 +02001361 return -EINVAL;
Jens Axboecbb7e572006-04-11 14:57:50 +02001362 if (copy_from_user(&offset, off_in, sizeof(loff_t)))
Jens Axboeb92ce552006-04-11 13:52:07 +02001363 return -EFAULT;
Jens Axboecbb7e572006-04-11 14:57:50 +02001364 off = &offset;
1365 } else
1366 off = &in->f_pos;
Ingo Molnar529565d2006-04-10 15:18:58 +02001367
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001368 ret = do_splice_to(in, off, opipe, len, flags);
Jens Axboea4514eb2006-04-19 15:57:05 +02001369
1370 if (off_in && copy_to_user(off_in, off, sizeof(loff_t)))
1371 ret = -EFAULT;
1372
1373 return ret;
Ingo Molnar529565d2006-04-10 15:18:58 +02001374 }
Jens Axboe5274f052006-03-30 15:15:30 +02001375
1376 return -EINVAL;
1377}
1378
Jens Axboe912d35f2006-04-26 10:59:21 +02001379/*
1380 * Map an iov into an array of pages and offset/length tupples. With the
1381 * partial_page structure, we can map several non-contiguous ranges into
1382 * our ones pages[] map instead of splitting that operation into pieces.
1383 * Could easily be exported as a generic helper for other users, in which
1384 * case one would probably want to add a 'max_nr_pages' parameter as well.
1385 */
1386static int get_iovec_page_array(const struct iovec __user *iov,
1387 unsigned int nr_vecs, struct page **pages,
Jens Axboe35f3d142010-05-20 10:43:18 +02001388 struct partial_page *partial, int aligned,
1389 unsigned int pipe_buffers)
Jens Axboe912d35f2006-04-26 10:59:21 +02001390{
1391 int buffers = 0, error = 0;
1392
Jens Axboe912d35f2006-04-26 10:59:21 +02001393 while (nr_vecs) {
1394 unsigned long off, npages;
Linus Torvalds75723952007-10-01 13:17:28 -07001395 struct iovec entry;
Jens Axboe912d35f2006-04-26 10:59:21 +02001396 void __user *base;
1397 size_t len;
1398 int i;
1399
Linus Torvalds75723952007-10-01 13:17:28 -07001400 error = -EFAULT;
Nick Pigginbc40d732008-07-25 19:45:26 -07001401 if (copy_from_user(&entry, iov, sizeof(entry)))
Jens Axboe912d35f2006-04-26 10:59:21 +02001402 break;
Linus Torvalds75723952007-10-01 13:17:28 -07001403
1404 base = entry.iov_base;
1405 len = entry.iov_len;
Jens Axboe912d35f2006-04-26 10:59:21 +02001406
1407 /*
1408 * Sanity check this iovec. 0 read succeeds.
1409 */
Linus Torvalds75723952007-10-01 13:17:28 -07001410 error = 0;
Jens Axboe912d35f2006-04-26 10:59:21 +02001411 if (unlikely(!len))
1412 break;
1413 error = -EFAULT;
Bastian Blank712a30e2008-02-10 16:47:57 +02001414 if (!access_ok(VERIFY_READ, base, len))
Jens Axboe912d35f2006-04-26 10:59:21 +02001415 break;
1416
1417 /*
1418 * Get this base offset and number of pages, then map
1419 * in the user pages.
1420 */
1421 off = (unsigned long) base & ~PAGE_MASK;
Jens Axboe7afa6fd2006-05-01 20:02:33 +02001422
1423 /*
1424 * If asked for alignment, the offset must be zero and the
1425 * length a multiple of the PAGE_SIZE.
1426 */
1427 error = -EINVAL;
1428 if (aligned && (off || len & ~PAGE_MASK))
1429 break;
1430
Jens Axboe912d35f2006-04-26 10:59:21 +02001431 npages = (off + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
Jens Axboe35f3d142010-05-20 10:43:18 +02001432 if (npages > pipe_buffers - buffers)
1433 npages = pipe_buffers - buffers;
Jens Axboe912d35f2006-04-26 10:59:21 +02001434
Nick Pigginbc40d732008-07-25 19:45:26 -07001435 error = get_user_pages_fast((unsigned long)base, npages,
1436 0, &pages[buffers]);
Jens Axboe912d35f2006-04-26 10:59:21 +02001437
1438 if (unlikely(error <= 0))
1439 break;
1440
1441 /*
1442 * Fill this contiguous range into the partial page map.
1443 */
1444 for (i = 0; i < error; i++) {
Jens Axboe75914892006-05-02 12:57:18 +02001445 const int plen = min_t(size_t, len, PAGE_SIZE - off);
Jens Axboe912d35f2006-04-26 10:59:21 +02001446
1447 partial[buffers].offset = off;
1448 partial[buffers].len = plen;
1449
1450 off = 0;
1451 len -= plen;
1452 buffers++;
1453 }
1454
1455 /*
1456 * We didn't complete this iov, stop here since it probably
1457 * means we have to move some of this into a pipe to
1458 * be able to continue.
1459 */
1460 if (len)
1461 break;
1462
1463 /*
1464 * Don't continue if we mapped fewer pages than we asked for,
1465 * or if we mapped the max number of pages that we have
1466 * room for.
1467 */
Jens Axboe35f3d142010-05-20 10:43:18 +02001468 if (error < npages || buffers == pipe_buffers)
Jens Axboe912d35f2006-04-26 10:59:21 +02001469 break;
1470
1471 nr_vecs--;
1472 iov++;
1473 }
1474
Jens Axboe912d35f2006-04-26 10:59:21 +02001475 if (buffers)
1476 return buffers;
1477
1478 return error;
1479}
1480
Jens Axboe6a14b902007-06-14 13:08:55 +02001481static int pipe_to_user(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
1482 struct splice_desc *sd)
1483{
1484 char *src;
1485 int ret;
1486
Jens Axboe6a14b902007-06-14 13:08:55 +02001487 /*
1488 * See if we can use the atomic maps, by prefaulting in the
1489 * pages and doing an atomic copy
1490 */
1491 if (!fault_in_pages_writeable(sd->u.userptr, sd->len)) {
1492 src = buf->ops->map(pipe, buf, 1);
1493 ret = __copy_to_user_inatomic(sd->u.userptr, src + buf->offset,
1494 sd->len);
1495 buf->ops->unmap(pipe, buf, src);
1496 if (!ret) {
1497 ret = sd->len;
1498 goto out;
1499 }
1500 }
1501
1502 /*
1503 * No dice, use slow non-atomic map and copy
1504 */
1505 src = buf->ops->map(pipe, buf, 0);
1506
1507 ret = sd->len;
1508 if (copy_to_user(sd->u.userptr, src + buf->offset, sd->len))
1509 ret = -EFAULT;
1510
Jens Axboe6866bef2007-10-16 10:01:29 +02001511 buf->ops->unmap(pipe, buf, src);
Jens Axboe6a14b902007-06-14 13:08:55 +02001512out:
1513 if (ret > 0)
1514 sd->u.userptr += ret;
Jens Axboe6a14b902007-06-14 13:08:55 +02001515 return ret;
1516}
1517
1518/*
1519 * For lack of a better implementation, implement vmsplice() to userspace
1520 * as a simple copy of the pipes pages to the user iov.
1521 */
1522static long vmsplice_to_user(struct file *file, const struct iovec __user *iov,
1523 unsigned long nr_segs, unsigned int flags)
1524{
1525 struct pipe_inode_info *pipe;
1526 struct splice_desc sd;
1527 ssize_t size;
1528 int error;
1529 long ret;
1530
Linus Torvalds71993e62010-11-28 13:56:09 -08001531 pipe = get_pipe_info(file);
Jens Axboe6a14b902007-06-14 13:08:55 +02001532 if (!pipe)
1533 return -EBADF;
1534
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001535 pipe_lock(pipe);
Jens Axboe6a14b902007-06-14 13:08:55 +02001536
1537 error = ret = 0;
1538 while (nr_segs) {
1539 void __user *base;
1540 size_t len;
1541
1542 /*
1543 * Get user address base and length for this iovec.
1544 */
1545 error = get_user(base, &iov->iov_base);
1546 if (unlikely(error))
1547 break;
1548 error = get_user(len, &iov->iov_len);
1549 if (unlikely(error))
1550 break;
1551
1552 /*
1553 * Sanity check this iovec. 0 read succeeds.
1554 */
1555 if (unlikely(!len))
1556 break;
1557 if (unlikely(!base)) {
1558 error = -EFAULT;
1559 break;
1560 }
1561
Jens Axboe88119302008-02-08 08:49:14 -08001562 if (unlikely(!access_ok(VERIFY_WRITE, base, len))) {
1563 error = -EFAULT;
1564 break;
1565 }
1566
Jens Axboe6a14b902007-06-14 13:08:55 +02001567 sd.len = 0;
1568 sd.total_len = len;
1569 sd.flags = flags;
1570 sd.u.userptr = base;
1571 sd.pos = 0;
1572
1573 size = __splice_from_pipe(pipe, &sd, pipe_to_user);
1574 if (size < 0) {
1575 if (!ret)
1576 ret = size;
1577
1578 break;
1579 }
1580
1581 ret += size;
1582
1583 if (size < len)
1584 break;
1585
1586 nr_segs--;
1587 iov++;
1588 }
1589
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001590 pipe_unlock(pipe);
Jens Axboe6a14b902007-06-14 13:08:55 +02001591
1592 if (!ret)
1593 ret = error;
1594
1595 return ret;
1596}
1597
Jens Axboe912d35f2006-04-26 10:59:21 +02001598/*
1599 * vmsplice splices a user address range into a pipe. It can be thought of
1600 * as splice-from-memory, where the regular splice is splice-from-file (or
1601 * to file). In both cases the output is a pipe, naturally.
Jens Axboe912d35f2006-04-26 10:59:21 +02001602 */
Jens Axboe6a14b902007-06-14 13:08:55 +02001603static long vmsplice_to_pipe(struct file *file, const struct iovec __user *iov,
1604 unsigned long nr_segs, unsigned int flags)
Jens Axboe912d35f2006-04-26 10:59:21 +02001605{
Jens Axboeddac0d32006-11-04 12:49:32 +01001606 struct pipe_inode_info *pipe;
Jens Axboe35f3d142010-05-20 10:43:18 +02001607 struct page *pages[PIPE_DEF_BUFFERS];
1608 struct partial_page partial[PIPE_DEF_BUFFERS];
Jens Axboe912d35f2006-04-26 10:59:21 +02001609 struct splice_pipe_desc spd = {
1610 .pages = pages,
1611 .partial = partial,
1612 .flags = flags,
1613 .ops = &user_page_pipe_buf_ops,
Jens Axboebbdfc2f2007-11-06 23:29:47 -08001614 .spd_release = spd_release_page,
Jens Axboe912d35f2006-04-26 10:59:21 +02001615 };
Jens Axboe35f3d142010-05-20 10:43:18 +02001616 long ret;
Jens Axboe912d35f2006-04-26 10:59:21 +02001617
Linus Torvalds71993e62010-11-28 13:56:09 -08001618 pipe = get_pipe_info(file);
Jens Axboeddac0d32006-11-04 12:49:32 +01001619 if (!pipe)
Jens Axboe912d35f2006-04-26 10:59:21 +02001620 return -EBADF;
Jens Axboe912d35f2006-04-26 10:59:21 +02001621
Jens Axboe35f3d142010-05-20 10:43:18 +02001622 if (splice_grow_spd(pipe, &spd))
1623 return -ENOMEM;
Jens Axboe912d35f2006-04-26 10:59:21 +02001624
Jens Axboe35f3d142010-05-20 10:43:18 +02001625 spd.nr_pages = get_iovec_page_array(iov, nr_segs, spd.pages,
1626 spd.partial, flags & SPLICE_F_GIFT,
1627 pipe->buffers);
1628 if (spd.nr_pages <= 0)
1629 ret = spd.nr_pages;
1630 else
1631 ret = splice_to_pipe(pipe, &spd);
1632
1633 splice_shrink_spd(pipe, &spd);
1634 return ret;
Jens Axboe912d35f2006-04-26 10:59:21 +02001635}
1636
Jens Axboe6a14b902007-06-14 13:08:55 +02001637/*
1638 * Note that vmsplice only really supports true splicing _from_ user memory
1639 * to a pipe, not the other way around. Splicing from user memory is a simple
1640 * operation that can be supported without any funky alignment restrictions
1641 * or nasty vm tricks. We simply map in the user memory and fill them into
1642 * a pipe. The reverse isn't quite as easy, though. There are two possible
1643 * solutions for that:
1644 *
1645 * - memcpy() the data internally, at which point we might as well just
1646 * do a regular read() on the buffer anyway.
1647 * - Lots of nasty vm tricks, that are neither fast nor flexible (it
1648 * has restriction limitations on both ends of the pipe).
1649 *
1650 * Currently we punt and implement it as a normal copy, see pipe_to_user().
1651 *
1652 */
Heiko Carstens836f92a2009-01-14 14:14:33 +01001653SYSCALL_DEFINE4(vmsplice, int, fd, const struct iovec __user *, iov,
1654 unsigned long, nr_segs, unsigned int, flags)
Jens Axboe912d35f2006-04-26 10:59:21 +02001655{
1656 struct file *file;
1657 long error;
1658 int fput;
1659
Jens Axboe6a14b902007-06-14 13:08:55 +02001660 if (unlikely(nr_segs > UIO_MAXIOV))
1661 return -EINVAL;
1662 else if (unlikely(!nr_segs))
1663 return 0;
1664
Jens Axboe912d35f2006-04-26 10:59:21 +02001665 error = -EBADF;
1666 file = fget_light(fd, &fput);
1667 if (file) {
1668 if (file->f_mode & FMODE_WRITE)
Jens Axboe6a14b902007-06-14 13:08:55 +02001669 error = vmsplice_to_pipe(file, iov, nr_segs, flags);
1670 else if (file->f_mode & FMODE_READ)
1671 error = vmsplice_to_user(file, iov, nr_segs, flags);
Jens Axboe912d35f2006-04-26 10:59:21 +02001672
1673 fput_light(file, fput);
1674 }
1675
1676 return error;
1677}
1678
Heiko Carstens836f92a2009-01-14 14:14:33 +01001679SYSCALL_DEFINE6(splice, int, fd_in, loff_t __user *, off_in,
1680 int, fd_out, loff_t __user *, off_out,
1681 size_t, len, unsigned int, flags)
Jens Axboe5274f052006-03-30 15:15:30 +02001682{
1683 long error;
1684 struct file *in, *out;
1685 int fput_in, fput_out;
1686
1687 if (unlikely(!len))
1688 return 0;
1689
1690 error = -EBADF;
Ingo Molnar529565d2006-04-10 15:18:58 +02001691 in = fget_light(fd_in, &fput_in);
Jens Axboe5274f052006-03-30 15:15:30 +02001692 if (in) {
1693 if (in->f_mode & FMODE_READ) {
Ingo Molnar529565d2006-04-10 15:18:58 +02001694 out = fget_light(fd_out, &fput_out);
Jens Axboe5274f052006-03-30 15:15:30 +02001695 if (out) {
1696 if (out->f_mode & FMODE_WRITE)
Ingo Molnar529565d2006-04-10 15:18:58 +02001697 error = do_splice(in, off_in,
1698 out, off_out,
1699 len, flags);
Jens Axboe5274f052006-03-30 15:15:30 +02001700 fput_light(out, fput_out);
1701 }
1702 }
1703
1704 fput_light(in, fput_in);
1705 }
1706
1707 return error;
1708}
Jens Axboe70524492006-04-11 15:51:17 +02001709
1710/*
Jens Axboeaadd06e2006-07-10 11:00:01 +02001711 * Make sure there's data to read. Wait for input if we can, otherwise
1712 * return an appropriate error.
1713 */
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001714static int ipipe_prep(struct pipe_inode_info *pipe, unsigned int flags)
Jens Axboeaadd06e2006-07-10 11:00:01 +02001715{
1716 int ret;
1717
1718 /*
1719 * Check ->nrbufs without the inode lock first. This function
1720 * is speculative anyways, so missing one is ok.
1721 */
1722 if (pipe->nrbufs)
1723 return 0;
1724
1725 ret = 0;
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001726 pipe_lock(pipe);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001727
1728 while (!pipe->nrbufs) {
1729 if (signal_pending(current)) {
1730 ret = -ERESTARTSYS;
1731 break;
1732 }
1733 if (!pipe->writers)
1734 break;
1735 if (!pipe->waiting_writers) {
1736 if (flags & SPLICE_F_NONBLOCK) {
1737 ret = -EAGAIN;
1738 break;
1739 }
1740 }
1741 pipe_wait(pipe);
1742 }
1743
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001744 pipe_unlock(pipe);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001745 return ret;
1746}
1747
1748/*
1749 * Make sure there's writeable room. Wait for room if we can, otherwise
1750 * return an appropriate error.
1751 */
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001752static int opipe_prep(struct pipe_inode_info *pipe, unsigned int flags)
Jens Axboeaadd06e2006-07-10 11:00:01 +02001753{
1754 int ret;
1755
1756 /*
1757 * Check ->nrbufs without the inode lock first. This function
1758 * is speculative anyways, so missing one is ok.
1759 */
Jens Axboe35f3d142010-05-20 10:43:18 +02001760 if (pipe->nrbufs < pipe->buffers)
Jens Axboeaadd06e2006-07-10 11:00:01 +02001761 return 0;
1762
1763 ret = 0;
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001764 pipe_lock(pipe);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001765
Jens Axboe35f3d142010-05-20 10:43:18 +02001766 while (pipe->nrbufs >= pipe->buffers) {
Jens Axboeaadd06e2006-07-10 11:00:01 +02001767 if (!pipe->readers) {
1768 send_sig(SIGPIPE, current, 0);
1769 ret = -EPIPE;
1770 break;
1771 }
1772 if (flags & SPLICE_F_NONBLOCK) {
1773 ret = -EAGAIN;
1774 break;
1775 }
1776 if (signal_pending(current)) {
1777 ret = -ERESTARTSYS;
1778 break;
1779 }
1780 pipe->waiting_writers++;
1781 pipe_wait(pipe);
1782 pipe->waiting_writers--;
1783 }
1784
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001785 pipe_unlock(pipe);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001786 return ret;
1787}
1788
1789/*
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001790 * Splice contents of ipipe to opipe.
1791 */
1792static int splice_pipe_to_pipe(struct pipe_inode_info *ipipe,
1793 struct pipe_inode_info *opipe,
1794 size_t len, unsigned int flags)
1795{
1796 struct pipe_buffer *ibuf, *obuf;
1797 int ret = 0, nbuf;
1798 bool input_wakeup = false;
1799
1800
1801retry:
1802 ret = ipipe_prep(ipipe, flags);
1803 if (ret)
1804 return ret;
1805
1806 ret = opipe_prep(opipe, flags);
1807 if (ret)
1808 return ret;
1809
1810 /*
1811 * Potential ABBA deadlock, work around it by ordering lock
1812 * grabbing by pipe info address. Otherwise two different processes
1813 * could deadlock (one doing tee from A -> B, the other from B -> A).
1814 */
1815 pipe_double_lock(ipipe, opipe);
1816
1817 do {
1818 if (!opipe->readers) {
1819 send_sig(SIGPIPE, current, 0);
1820 if (!ret)
1821 ret = -EPIPE;
1822 break;
1823 }
1824
1825 if (!ipipe->nrbufs && !ipipe->writers)
1826 break;
1827
1828 /*
1829 * Cannot make any progress, because either the input
1830 * pipe is empty or the output pipe is full.
1831 */
Jens Axboe35f3d142010-05-20 10:43:18 +02001832 if (!ipipe->nrbufs || opipe->nrbufs >= opipe->buffers) {
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001833 /* Already processed some buffers, break */
1834 if (ret)
1835 break;
1836
1837 if (flags & SPLICE_F_NONBLOCK) {
1838 ret = -EAGAIN;
1839 break;
1840 }
1841
1842 /*
1843 * We raced with another reader/writer and haven't
1844 * managed to process any buffers. A zero return
1845 * value means EOF, so retry instead.
1846 */
1847 pipe_unlock(ipipe);
1848 pipe_unlock(opipe);
1849 goto retry;
1850 }
1851
1852 ibuf = ipipe->bufs + ipipe->curbuf;
Jens Axboe35f3d142010-05-20 10:43:18 +02001853 nbuf = (opipe->curbuf + opipe->nrbufs) & (opipe->buffers - 1);
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001854 obuf = opipe->bufs + nbuf;
1855
1856 if (len >= ibuf->len) {
1857 /*
1858 * Simply move the whole buffer from ipipe to opipe
1859 */
1860 *obuf = *ibuf;
1861 ibuf->ops = NULL;
1862 opipe->nrbufs++;
Jens Axboe35f3d142010-05-20 10:43:18 +02001863 ipipe->curbuf = (ipipe->curbuf + 1) & (ipipe->buffers - 1);
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001864 ipipe->nrbufs--;
1865 input_wakeup = true;
1866 } else {
1867 /*
1868 * Get a reference to this pipe buffer,
1869 * so we can copy the contents over.
1870 */
1871 ibuf->ops->get(ipipe, ibuf);
1872 *obuf = *ibuf;
1873
1874 /*
1875 * Don't inherit the gift flag, we need to
1876 * prevent multiple steals of this page.
1877 */
1878 obuf->flags &= ~PIPE_BUF_FLAG_GIFT;
1879
1880 obuf->len = len;
1881 opipe->nrbufs++;
1882 ibuf->offset += obuf->len;
1883 ibuf->len -= obuf->len;
1884 }
1885 ret += obuf->len;
1886 len -= obuf->len;
1887 } while (len);
1888
1889 pipe_unlock(ipipe);
1890 pipe_unlock(opipe);
1891
1892 /*
1893 * If we put data in the output pipe, wakeup any potential readers.
1894 */
1895 if (ret > 0) {
1896 smp_mb();
1897 if (waitqueue_active(&opipe->wait))
1898 wake_up_interruptible(&opipe->wait);
1899 kill_fasync(&opipe->fasync_readers, SIGIO, POLL_IN);
1900 }
1901 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 Axboe2a27250e2006-04-19 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 Axboe2a27250e2006-04-19 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 */
1979 if (ret > 0) {
Jens Axboe70524492006-04-11 15:51:17 +02001980 smp_mb();
1981 if (waitqueue_active(&opipe->wait))
1982 wake_up_interruptible(&opipe->wait);
1983 kill_fasync(&opipe->fasync_readers, SIGIO, POLL_IN);
1984 }
1985
1986 return ret;
1987}
1988
1989/*
1990 * This is a tee(1) implementation that works on pipes. It doesn't copy
1991 * any data, it simply references the 'in' pages on the 'out' pipe.
1992 * The 'flags' used are the SPLICE_F_* variants, currently the only
1993 * applicable one is SPLICE_F_NONBLOCK.
1994 */
1995static long do_tee(struct file *in, struct file *out, size_t len,
1996 unsigned int flags)
1997{
Linus Torvalds71993e62010-11-28 13:56:09 -08001998 struct pipe_inode_info *ipipe = get_pipe_info(in);
1999 struct pipe_inode_info *opipe = get_pipe_info(out);
Jens Axboeaadd06e2006-07-10 11:00:01 +02002000 int ret = -EINVAL;
Jens Axboe70524492006-04-11 15:51:17 +02002001
2002 /*
Jens Axboeaadd06e2006-07-10 11:00:01 +02002003 * Duplicate the contents of ipipe to opipe without actually
2004 * copying the data.
Jens Axboe70524492006-04-11 15:51:17 +02002005 */
Jens Axboeaadd06e2006-07-10 11:00:01 +02002006 if (ipipe && opipe && ipipe != opipe) {
2007 /*
2008 * Keep going, unless we encounter an error. The ipipe/opipe
2009 * ordering doesn't really matter.
2010 */
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02002011 ret = ipipe_prep(ipipe, flags);
Jens Axboeaadd06e2006-07-10 11:00:01 +02002012 if (!ret) {
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02002013 ret = opipe_prep(opipe, flags);
Jens Axboe02cf01a2008-02-20 10:34:51 +01002014 if (!ret)
Jens Axboeaadd06e2006-07-10 11:00:01 +02002015 ret = link_pipe(ipipe, opipe, len, flags);
Jens Axboeaadd06e2006-07-10 11:00:01 +02002016 }
2017 }
Jens Axboe70524492006-04-11 15:51:17 +02002018
Jens Axboeaadd06e2006-07-10 11:00:01 +02002019 return ret;
Jens Axboe70524492006-04-11 15:51:17 +02002020}
2021
Heiko Carstens836f92a2009-01-14 14:14:33 +01002022SYSCALL_DEFINE4(tee, int, fdin, int, fdout, size_t, len, unsigned int, flags)
Jens Axboe70524492006-04-11 15:51:17 +02002023{
2024 struct file *in;
2025 int error, fput_in;
2026
2027 if (unlikely(!len))
2028 return 0;
2029
2030 error = -EBADF;
2031 in = fget_light(fdin, &fput_in);
2032 if (in) {
2033 if (in->f_mode & FMODE_READ) {
2034 int fput_out;
2035 struct file *out = fget_light(fdout, &fput_out);
2036
2037 if (out) {
2038 if (out->f_mode & FMODE_WRITE)
2039 error = do_tee(in, out, len, flags);
2040 fput_light(out, fput_out);
2041 }
2042 }
2043 fput_light(in, fput_in);
2044 }
2045
2046 return error;
2047}