blob: 740e6b9faf7ab2e10e0ebfc68725d0d05e0b547f [file] [log] [blame]
Jens Axboe5274f052006-03-30 15:15:30 +02001/*
2 * "splice": joining two ropes together by interweaving their strands.
3 *
4 * This is the "extended pipe" functionality, where a pipe is used as
5 * an arbitrary in-memory buffer. Think of a pipe as a small kernel
6 * buffer that you can use to transfer data from one end to the other.
7 *
8 * The traditional unix read/write is extended with a "splice()" operation
9 * that transfers data buffers to or from a pipe buffer.
10 *
11 * Named by Larry McVoy, original implementation from Linus, extended by
Jens Axboec2058e02006-04-11 13:56:34 +020012 * Jens to support splicing to files, network, direct splicing, etc and
13 * fixing lots of bugs.
Jens Axboe5274f052006-03-30 15:15:30 +020014 *
Jens Axboe0fe23472006-09-04 15:41:16 +020015 * Copyright (C) 2005-2006 Jens Axboe <axboe@kernel.dk>
Jens Axboec2058e02006-04-11 13:56:34 +020016 * Copyright (C) 2005-2006 Linus Torvalds <torvalds@osdl.org>
17 * Copyright (C) 2006 Ingo Molnar <mingo@elte.hu>
Jens Axboe5274f052006-03-30 15:15:30 +020018 *
19 */
20#include <linux/fs.h>
21#include <linux/file.h>
22#include <linux/pagemap.h>
Jens Axboed6b29d72007-06-04 09:59:47 +020023#include <linux/splice.h>
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -080024#include <linux/memcontrol.h>
Jens Axboe5274f052006-03-30 15:15:30 +020025#include <linux/mm_inline.h>
Jens Axboe5abc97a2006-03-30 15:16:46 +020026#include <linux/swap.h>
Jens Axboe4f6f0bd2006-04-02 23:04:46 +020027#include <linux/writeback.h>
28#include <linux/buffer_head.h>
Jeff Garzika0f06782006-03-30 23:06:13 -050029#include <linux/module.h>
Jens Axboe4f6f0bd2006-04-02 23:04:46 +020030#include <linux/syscalls.h>
Jens Axboe912d35f2006-04-26 10:59:21 +020031#include <linux/uio.h>
James Morris29ce2052007-07-13 11:44:32 +020032#include <linux/security.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090033#include <linux/gfp.h>
Jens Axboe5274f052006-03-30 15:15:30 +020034
Jens Axboe83f91352006-04-02 23:05:09 +020035/*
36 * Attempt to steal a page from a pipe buffer. This should perhaps go into
37 * a vm helper function, it's already simplified quite a bit by the
38 * addition of remove_mapping(). If success is returned, the caller may
39 * attempt to reuse this page for another destination.
40 */
Jens Axboe76ad4d12006-05-03 10:41:33 +020041static int page_cache_pipe_buf_steal(struct pipe_inode_info *pipe,
Jens Axboe5abc97a2006-03-30 15:16:46 +020042 struct pipe_buffer *buf)
43{
44 struct page *page = buf->page;
Jens Axboe9e94cd42006-06-20 15:01:12 +020045 struct address_space *mapping;
Jens Axboe5abc97a2006-03-30 15:16:46 +020046
Jens Axboe9e0267c2006-04-19 15:57:31 +020047 lock_page(page);
48
Jens Axboe9e94cd42006-06-20 15:01:12 +020049 mapping = page_mapping(page);
50 if (mapping) {
51 WARN_ON(!PageUptodate(page));
Jens Axboe5abc97a2006-03-30 15:16:46 +020052
Jens Axboe9e94cd42006-06-20 15:01:12 +020053 /*
54 * At least for ext2 with nobh option, we need to wait on
55 * writeback completing on this page, since we'll remove it
56 * from the pagecache. Otherwise truncate wont wait on the
57 * page, allowing the disk blocks to be reused by someone else
58 * before we actually wrote our data to them. fs corruption
59 * ensues.
60 */
61 wait_on_page_writeback(page);
Jens Axboead8d6f02006-04-02 23:10:32 +020062
David Howells266cf652009-04-03 16:42:36 +010063 if (page_has_private(page) &&
64 !try_to_release_page(page, GFP_KERNEL))
Jens Axboeca39d652008-05-20 21:27:41 +020065 goto out_unlock;
Jens Axboe4f6f0bd2006-04-02 23:04:46 +020066
Jens Axboe9e94cd42006-06-20 15:01:12 +020067 /*
68 * If we succeeded in removing the mapping, set LRU flag
69 * and return good.
70 */
71 if (remove_mapping(mapping, page)) {
72 buf->flags |= PIPE_BUF_FLAG_LRU;
73 return 0;
74 }
Jens Axboe9e0267c2006-04-19 15:57:31 +020075 }
Jens Axboe5abc97a2006-03-30 15:16:46 +020076
Jens Axboe9e94cd42006-06-20 15:01:12 +020077 /*
78 * Raced with truncate or failed to remove page from current
79 * address space, unlock and return failure.
80 */
Jens Axboeca39d652008-05-20 21:27:41 +020081out_unlock:
Jens Axboe9e94cd42006-06-20 15:01:12 +020082 unlock_page(page);
83 return 1;
Jens Axboe5abc97a2006-03-30 15:16:46 +020084}
85
Jens Axboe76ad4d12006-05-03 10:41:33 +020086static void page_cache_pipe_buf_release(struct pipe_inode_info *pipe,
Jens Axboe5274f052006-03-30 15:15:30 +020087 struct pipe_buffer *buf)
88{
89 page_cache_release(buf->page);
Jens Axboe14328732006-05-03 10:35:26 +020090 buf->flags &= ~PIPE_BUF_FLAG_LRU;
Jens Axboe5274f052006-03-30 15:15:30 +020091}
92
Jens Axboe08457182007-06-12 20:51:32 +020093/*
94 * Check whether the contents of buf is OK to access. Since the content
95 * is a page cache page, IO may be in flight.
96 */
Jens Axboecac36bb2007-06-14 13:10:48 +020097static int page_cache_pipe_buf_confirm(struct pipe_inode_info *pipe,
98 struct pipe_buffer *buf)
Jens Axboe5274f052006-03-30 15:15:30 +020099{
100 struct page *page = buf->page;
Jens Axboe49d0b212006-04-10 09:04:41 +0200101 int err;
Jens Axboe5274f052006-03-30 15:15:30 +0200102
103 if (!PageUptodate(page)) {
Jens Axboe49d0b212006-04-10 09:04:41 +0200104 lock_page(page);
105
106 /*
107 * Page got truncated/unhashed. This will cause a 0-byte
Ingo Molnar73d62d82006-04-11 13:57:21 +0200108 * splice, if this is the first page.
Jens Axboe49d0b212006-04-10 09:04:41 +0200109 */
110 if (!page->mapping) {
111 err = -ENODATA;
112 goto error;
113 }
114
115 /*
Ingo Molnar73d62d82006-04-11 13:57:21 +0200116 * Uh oh, read-error from disk.
Jens Axboe49d0b212006-04-10 09:04:41 +0200117 */
118 if (!PageUptodate(page)) {
119 err = -EIO;
120 goto error;
121 }
122
123 /*
Jens Axboef84d7512006-05-01 19:59:03 +0200124 * Page is ok afterall, we are done.
Jens Axboe49d0b212006-04-10 09:04:41 +0200125 */
Jens Axboe5274f052006-03-30 15:15:30 +0200126 unlock_page(page);
Jens Axboe5274f052006-03-30 15:15:30 +0200127 }
128
Jens Axboef84d7512006-05-01 19:59:03 +0200129 return 0;
Jens Axboe49d0b212006-04-10 09:04:41 +0200130error:
131 unlock_page(page);
Jens Axboef84d7512006-05-01 19:59:03 +0200132 return err;
Jens Axboe70524492006-04-11 15:51:17 +0200133}
134
Eric Dumazetd4c3cca2006-12-13 00:34:04 -0800135static const struct pipe_buf_operations page_cache_pipe_buf_ops = {
Jens Axboe5274f052006-03-30 15:15:30 +0200136 .can_merge = 0,
Jens Axboef84d7512006-05-01 19:59:03 +0200137 .map = generic_pipe_buf_map,
138 .unmap = generic_pipe_buf_unmap,
Jens Axboecac36bb2007-06-14 13:10:48 +0200139 .confirm = page_cache_pipe_buf_confirm,
Jens Axboe5274f052006-03-30 15:15:30 +0200140 .release = page_cache_pipe_buf_release,
Jens Axboe5abc97a2006-03-30 15:16:46 +0200141 .steal = page_cache_pipe_buf_steal,
Jens Axboef84d7512006-05-01 19:59:03 +0200142 .get = generic_pipe_buf_get,
Jens Axboe5274f052006-03-30 15:15:30 +0200143};
144
Jens Axboe912d35f2006-04-26 10:59:21 +0200145static int user_page_pipe_buf_steal(struct pipe_inode_info *pipe,
146 struct pipe_buffer *buf)
147{
Jens Axboe7afa6fd2006-05-01 20:02:33 +0200148 if (!(buf->flags & PIPE_BUF_FLAG_GIFT))
149 return 1;
150
Jens Axboe14328732006-05-03 10:35:26 +0200151 buf->flags |= PIPE_BUF_FLAG_LRU;
Jens Axboe330ab712006-05-02 15:29:57 +0200152 return generic_pipe_buf_steal(pipe, buf);
Jens Axboe912d35f2006-04-26 10:59:21 +0200153}
154
Eric Dumazetd4c3cca2006-12-13 00:34:04 -0800155static const struct pipe_buf_operations user_page_pipe_buf_ops = {
Jens Axboe912d35f2006-04-26 10:59:21 +0200156 .can_merge = 0,
Jens Axboef84d7512006-05-01 19:59:03 +0200157 .map = generic_pipe_buf_map,
158 .unmap = generic_pipe_buf_unmap,
Jens Axboecac36bb2007-06-14 13:10:48 +0200159 .confirm = generic_pipe_buf_confirm,
Jens Axboe912d35f2006-04-26 10:59:21 +0200160 .release = page_cache_pipe_buf_release,
161 .steal = user_page_pipe_buf_steal,
Jens Axboef84d7512006-05-01 19:59:03 +0200162 .get = generic_pipe_buf_get,
Jens Axboe912d35f2006-04-26 10:59:21 +0200163};
164
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 Wu431a4822007-07-19 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 Wu431a4822007-07-19 01:48:05 -0700335 * readahead/allocate the rest and fill in the holes.
Jens Axboeeb207962006-04-27 11:05:22 +0200336 */
Fengguang Wu431a4822007-07-19 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 Wu431a4822007-07-19 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)) {
Jens Axboec4f895c2006-04-19 15:56:12 +0200402 /*
403 * If in nonblock mode then dont block on waiting
404 * for an in-flight io page
405 */
Fengguang Wu9ae9d682007-05-08 08:44:36 +0200406 if (flags & SPLICE_F_NONBLOCK) {
Nick Piggin529ae9a2008-08-02 12:01:03 +0200407 if (!trylock_page(page)) {
Jens Axboe8191ecd2008-04-10 08:24:25 +0200408 error = -EAGAIN;
Fengguang Wu9ae9d682007-05-08 08:44:36 +0200409 break;
Jens Axboe8191ecd2008-04-10 08:24:25 +0200410 }
Fengguang Wu9ae9d682007-05-08 08:44:36 +0200411 } else
412 lock_page(page);
Jens Axboe7480a902006-04-11 13:52:47 +0200413
414 /*
Miklos Szeredi32502b82008-07-04 09:35:17 +0200415 * Page was truncated, or invalidated by the
416 * filesystem. Redo the find/create, but this time the
417 * page is kept locked, so there's no chance of another
418 * race with truncate/invalidate.
Jens Axboe7480a902006-04-11 13:52:47 +0200419 */
420 if (!page->mapping) {
421 unlock_page(page);
Miklos Szeredi32502b82008-07-04 09:35:17 +0200422 page = find_or_create_page(mapping, index,
423 mapping_gfp_mask(mapping));
424
425 if (!page) {
426 error = -ENOMEM;
427 break;
428 }
Jens Axboe35f3d142010-05-20 10:43:18 +0200429 page_cache_release(spd.pages[page_nr]);
430 spd.pages[page_nr] = page;
Jens Axboe7480a902006-04-11 13:52:47 +0200431 }
432 /*
433 * page was already under io and is now done, great
434 */
435 if (PageUptodate(page)) {
436 unlock_page(page);
437 goto fill_it;
438 }
439
Jens Axboe7480a902006-04-11 13:52:47 +0200440 /*
441 * need to read in the page
442 */
443 error = mapping->a_ops->readpage(in, page);
Jens Axboe7480a902006-04-11 13:52:47 +0200444 if (unlikely(error)) {
Jens Axboeeb207962006-04-27 11:05:22 +0200445 /*
446 * We really should re-lookup the page here,
447 * but it complicates things a lot. Instead
448 * lets just do what we already stored, and
449 * we'll get it the next time we are called.
450 */
Jens Axboe7480a902006-04-11 13:52:47 +0200451 if (error == AOP_TRUNCATED_PAGE)
Jens Axboeeb207962006-04-27 11:05:22 +0200452 error = 0;
453
Jens Axboe7480a902006-04-11 13:52:47 +0200454 break;
455 }
Jens Axboe620a3242007-06-07 09:39:42 +0200456 }
457fill_it:
458 /*
459 * i_size must be checked after PageUptodate.
460 */
461 isize = i_size_read(mapping->host);
462 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
463 if (unlikely(!isize || index > end_index))
464 break;
465
466 /*
467 * if this is the last page, see if we need to shrink
468 * the length and stop
469 */
470 if (end_index == index) {
471 unsigned int plen;
Jens Axboe91ad66e2006-04-19 15:55:10 +0200472
473 /*
Jens Axboe620a3242007-06-07 09:39:42 +0200474 * max good bytes in this page
Jens Axboe91ad66e2006-04-19 15:55:10 +0200475 */
Jens Axboe620a3242007-06-07 09:39:42 +0200476 plen = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
477 if (plen <= loff)
Jens Axboe91ad66e2006-04-19 15:55:10 +0200478 break;
Jens Axboe91ad66e2006-04-19 15:55:10 +0200479
480 /*
Jens Axboe620a3242007-06-07 09:39:42 +0200481 * force quit after adding this page
Jens Axboe91ad66e2006-04-19 15:55:10 +0200482 */
Jens Axboe620a3242007-06-07 09:39:42 +0200483 this_len = min(this_len, plen - loff);
484 len = this_len;
Jens Axboe7480a902006-04-11 13:52:47 +0200485 }
Jens Axboe620a3242007-06-07 09:39:42 +0200486
Jens Axboe35f3d142010-05-20 10:43:18 +0200487 spd.partial[page_nr].offset = loff;
488 spd.partial[page_nr].len = this_len;
Jens Axboe82aa5d62006-04-20 13:05:48 +0200489 len -= this_len;
Jens Axboe91ad66e2006-04-19 15:55:10 +0200490 loff = 0;
Jens Axboeeb207962006-04-27 11:05:22 +0200491 spd.nr_pages++;
492 index++;
Jens Axboe5274f052006-03-30 15:15:30 +0200493 }
494
Jens Axboeeb207962006-04-27 11:05:22 +0200495 /*
Hugh Dickins475ecad2007-06-07 09:36:00 +0200496 * Release any pages at the end, if we quit early. 'page_nr' is how far
Jens Axboeeb207962006-04-27 11:05:22 +0200497 * we got, 'nr_pages' is how many pages are in the map.
498 */
499 while (page_nr < nr_pages)
Jens Axboe35f3d142010-05-20 10:43:18 +0200500 page_cache_release(spd.pages[page_nr++]);
Fengguang Wuf4e6b492007-10-16 01:24:33 -0700501 in->f_ra.prev_pos = (loff_t)index << PAGE_CACHE_SHIFT;
Jens Axboeeb207962006-04-27 11:05:22 +0200502
Jens Axboe912d35f2006-04-26 10:59:21 +0200503 if (spd.nr_pages)
Jens Axboe35f3d142010-05-20 10:43:18 +0200504 error = splice_to_pipe(pipe, &spd);
Jens Axboe5274f052006-03-30 15:15:30 +0200505
Jens Axboe35f3d142010-05-20 10:43:18 +0200506 splice_shrink_spd(pipe, &spd);
Jens Axboe7480a902006-04-11 13:52:47 +0200507 return error;
Jens Axboe5274f052006-03-30 15:15:30 +0200508}
509
Jens Axboe83f91352006-04-02 23:05:09 +0200510/**
511 * generic_file_splice_read - splice data from file to a pipe
512 * @in: file to splice from
Jens Axboe932cc6d2007-06-21 13:10:21 +0200513 * @ppos: position in @in
Jens Axboe83f91352006-04-02 23:05:09 +0200514 * @pipe: pipe to splice to
515 * @len: number of bytes to splice
516 * @flags: splice modifier flags
517 *
Jens Axboe932cc6d2007-06-21 13:10:21 +0200518 * Description:
519 * Will read pages from given file and fill them into a pipe. Can be
520 * used as long as the address_space operations for the source implements
521 * a readpage() hook.
522 *
Jens Axboe83f91352006-04-02 23:05:09 +0200523 */
Jens Axboecbb7e572006-04-11 14:57:50 +0200524ssize_t generic_file_splice_read(struct file *in, loff_t *ppos,
525 struct pipe_inode_info *pipe, size_t len,
526 unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +0200527{
Jens Axboed366d3982007-06-01 14:54:11 +0200528 loff_t isize, left;
Jens Axboe8191ecd2008-04-10 08:24:25 +0200529 int ret;
Jens Axboed366d3982007-06-01 14:54:11 +0200530
531 isize = i_size_read(in->f_mapping->host);
532 if (unlikely(*ppos >= isize))
533 return 0;
534
535 left = isize - *ppos;
536 if (unlikely(left < len))
537 len = left;
Jens Axboe5274f052006-03-30 15:15:30 +0200538
Jens Axboe8191ecd2008-04-10 08:24:25 +0200539 ret = __generic_file_splice_read(in, ppos, pipe, len, flags);
Miklos Szeredi723590e2009-08-15 08:43:22 +0200540 if (ret > 0) {
Jens Axboecbb7e572006-04-11 14:57:50 +0200541 *ppos += ret;
Miklos Szeredi723590e2009-08-15 08:43:22 +0200542 file_accessed(in);
543 }
Jens Axboe5274f052006-03-30 15:15:30 +0200544
545 return ret;
546}
Jens Axboe059a8f32006-04-02 23:06:05 +0200547EXPORT_SYMBOL(generic_file_splice_read);
548
Miklos Szeredi68181732009-05-07 15:37:36 +0200549static const struct pipe_buf_operations default_pipe_buf_ops = {
550 .can_merge = 0,
551 .map = generic_pipe_buf_map,
552 .unmap = generic_pipe_buf_unmap,
553 .confirm = generic_pipe_buf_confirm,
554 .release = generic_pipe_buf_release,
555 .steal = generic_pipe_buf_steal,
556 .get = generic_pipe_buf_get,
557};
558
559static ssize_t kernel_readv(struct file *file, const struct iovec *vec,
560 unsigned long vlen, loff_t offset)
561{
562 mm_segment_t old_fs;
563 loff_t pos = offset;
564 ssize_t res;
565
566 old_fs = get_fs();
567 set_fs(get_ds());
568 /* The cast to a user pointer is valid due to the set_fs() */
569 res = vfs_readv(file, (const struct iovec __user *)vec, vlen, &pos);
570 set_fs(old_fs);
571
572 return res;
573}
574
Miklos Szeredib2858d72009-05-19 11:37:46 +0200575static ssize_t kernel_write(struct file *file, const char *buf, size_t count,
576 loff_t pos)
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +0200577{
578 mm_segment_t old_fs;
579 ssize_t res;
580
581 old_fs = get_fs();
582 set_fs(get_ds());
583 /* The cast to a user pointer is valid due to the set_fs() */
Miklos Szeredib2858d72009-05-19 11:37:46 +0200584 res = vfs_write(file, (const char __user *)buf, count, &pos);
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +0200585 set_fs(old_fs);
586
587 return res;
588}
589
Miklos Szeredi68181732009-05-07 15:37:36 +0200590ssize_t default_file_splice_read(struct file *in, loff_t *ppos,
591 struct pipe_inode_info *pipe, size_t len,
592 unsigned int flags)
593{
594 unsigned int nr_pages;
595 unsigned int nr_freed;
596 size_t offset;
Jens Axboe35f3d142010-05-20 10:43:18 +0200597 struct page *pages[PIPE_DEF_BUFFERS];
598 struct partial_page partial[PIPE_DEF_BUFFERS];
599 struct iovec *vec, __vec[PIPE_DEF_BUFFERS];
Miklos Szeredi68181732009-05-07 15:37:36 +0200600 pgoff_t index;
601 ssize_t res;
602 size_t this_len;
603 int error;
604 int i;
605 struct splice_pipe_desc spd = {
606 .pages = pages,
607 .partial = partial,
608 .flags = flags,
609 .ops = &default_pipe_buf_ops,
610 .spd_release = spd_release_page,
611 };
612
Jens Axboe35f3d142010-05-20 10:43:18 +0200613 if (splice_grow_spd(pipe, &spd))
614 return -ENOMEM;
615
616 res = -ENOMEM;
617 vec = __vec;
618 if (pipe->buffers > PIPE_DEF_BUFFERS) {
619 vec = kmalloc(pipe->buffers * sizeof(struct iovec), GFP_KERNEL);
620 if (!vec)
621 goto shrink_ret;
622 }
623
Miklos Szeredi68181732009-05-07 15:37:36 +0200624 index = *ppos >> PAGE_CACHE_SHIFT;
625 offset = *ppos & ~PAGE_CACHE_MASK;
626 nr_pages = (len + offset + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
627
Jens Axboe35f3d142010-05-20 10:43:18 +0200628 for (i = 0; i < nr_pages && i < pipe->buffers && len; i++) {
Miklos Szeredi68181732009-05-07 15:37:36 +0200629 struct page *page;
630
Jens Axboe4f231222009-05-13 08:35:35 +0200631 page = alloc_page(GFP_USER);
Miklos Szeredi68181732009-05-07 15:37:36 +0200632 error = -ENOMEM;
633 if (!page)
634 goto err;
635
636 this_len = min_t(size_t, len, PAGE_CACHE_SIZE - offset);
Jens Axboe4f231222009-05-13 08:35:35 +0200637 vec[i].iov_base = (void __user *) page_address(page);
Miklos Szeredi68181732009-05-07 15:37:36 +0200638 vec[i].iov_len = this_len;
Jens Axboe35f3d142010-05-20 10:43:18 +0200639 spd.pages[i] = page;
Miklos Szeredi68181732009-05-07 15:37:36 +0200640 spd.nr_pages++;
641 len -= this_len;
642 offset = 0;
643 }
644
645 res = kernel_readv(in, vec, spd.nr_pages, *ppos);
Andrew Morton77f6bf52009-05-14 09:49:44 +0200646 if (res < 0) {
647 error = res;
Miklos Szeredi68181732009-05-07 15:37:36 +0200648 goto err;
Andrew Morton77f6bf52009-05-14 09:49:44 +0200649 }
Miklos Szeredi68181732009-05-07 15:37:36 +0200650
651 error = 0;
652 if (!res)
653 goto err;
654
655 nr_freed = 0;
656 for (i = 0; i < spd.nr_pages; i++) {
Miklos Szeredi68181732009-05-07 15:37:36 +0200657 this_len = min_t(size_t, vec[i].iov_len, res);
Jens Axboe35f3d142010-05-20 10:43:18 +0200658 spd.partial[i].offset = 0;
659 spd.partial[i].len = this_len;
Miklos Szeredi68181732009-05-07 15:37:36 +0200660 if (!this_len) {
Jens Axboe35f3d142010-05-20 10:43:18 +0200661 __free_page(spd.pages[i]);
662 spd.pages[i] = NULL;
Miklos Szeredi68181732009-05-07 15:37:36 +0200663 nr_freed++;
664 }
665 res -= this_len;
666 }
667 spd.nr_pages -= nr_freed;
668
669 res = splice_to_pipe(pipe, &spd);
670 if (res > 0)
671 *ppos += res;
672
Jens Axboe35f3d142010-05-20 10:43:18 +0200673shrink_ret:
674 if (vec != __vec)
675 kfree(vec);
676 splice_shrink_spd(pipe, &spd);
Miklos Szeredi68181732009-05-07 15:37:36 +0200677 return res;
678
679err:
Jens Axboe4f231222009-05-13 08:35:35 +0200680 for (i = 0; i < spd.nr_pages; i++)
Jens Axboe35f3d142010-05-20 10:43:18 +0200681 __free_page(spd.pages[i]);
Jens Axboe4f231222009-05-13 08:35:35 +0200682
Jens Axboe35f3d142010-05-20 10:43:18 +0200683 res = error;
684 goto shrink_ret;
Miklos Szeredi68181732009-05-07 15:37:36 +0200685}
686EXPORT_SYMBOL(default_file_splice_read);
687
Jens Axboe5274f052006-03-30 15:15:30 +0200688/*
Jens Axboe4f6f0bd2006-04-02 23:04:46 +0200689 * Send 'sd->len' bytes to socket from 'sd->file' at position 'sd->pos'
Jens Axboe016b6612006-04-25 15:42:00 +0200690 * using sendpage(). Return the number of bytes sent.
Jens Axboe5274f052006-03-30 15:15:30 +0200691 */
Jens Axboe76ad4d12006-05-03 10:41:33 +0200692static int pipe_to_sendpage(struct pipe_inode_info *pipe,
Jens Axboe5274f052006-03-30 15:15:30 +0200693 struct pipe_buffer *buf, struct splice_desc *sd)
694{
Jens Axboe6a14b902007-06-14 13:08:55 +0200695 struct file *file = sd->u.file;
Jens Axboe5274f052006-03-30 15:15:30 +0200696 loff_t pos = sd->pos;
Jens Axboef84d7512006-05-01 19:59:03 +0200697 int ret, more;
Jens Axboe5274f052006-03-30 15:15:30 +0200698
Jens Axboecac36bb2007-06-14 13:10:48 +0200699 ret = buf->ops->confirm(pipe, buf);
Jens Axboef84d7512006-05-01 19:59:03 +0200700 if (!ret) {
701 more = (sd->flags & SPLICE_F_MORE) || sd->len < sd->total_len;
Changli Gaocc56f7d2009-11-04 09:09:52 +0100702 if (file->f_op && file->f_op->sendpage)
703 ret = file->f_op->sendpage(file, buf->page, buf->offset,
704 sd->len, &pos, more);
705 else
706 ret = -EINVAL;
Jens Axboef84d7512006-05-01 19:59:03 +0200707 }
Jens Axboe5274f052006-03-30 15:15:30 +0200708
Jens Axboe016b6612006-04-25 15:42:00 +0200709 return ret;
Jens Axboe5274f052006-03-30 15:15:30 +0200710}
711
712/*
713 * This is a little more tricky than the file -> pipe splicing. There are
714 * basically three cases:
715 *
716 * - Destination page already exists in the address space and there
717 * are users of it. For that case we have no other option that
718 * copying the data. Tough luck.
719 * - Destination page already exists in the address space, but there
720 * are no users of it. Make sure it's uptodate, then drop it. Fall
721 * through to last case.
722 * - Destination page does not exist, we can add the pipe page to
723 * the page cache and avoid the copy.
724 *
Jens Axboe83f91352006-04-02 23:05:09 +0200725 * If asked to move pages to the output file (SPLICE_F_MOVE is set in
726 * sd->flags), we attempt to migrate pages from the pipe to the output
727 * file address space page cache. This is possible if no one else has
728 * the pipe page referenced outside of the pipe and page cache. If
729 * SPLICE_F_MOVE isn't set, or we cannot move the page, we simply create
730 * a new page in the output file page cache and fill/dirty that.
Jens Axboe5274f052006-03-30 15:15:30 +0200731 */
Miklos Szeredi328eaab2009-04-14 19:48:39 +0200732int pipe_to_file(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
733 struct splice_desc *sd)
Jens Axboe5274f052006-03-30 15:15:30 +0200734{
Jens Axboe6a14b902007-06-14 13:08:55 +0200735 struct file *file = sd->u.file;
Jens Axboe5274f052006-03-30 15:15:30 +0200736 struct address_space *mapping = file->f_mapping;
Jens Axboe016b6612006-04-25 15:42:00 +0200737 unsigned int offset, this_len;
Jens Axboe5274f052006-03-30 15:15:30 +0200738 struct page *page;
Nick Pigginafddba42007-10-16 01:25:01 -0700739 void *fsdata;
Jens Axboe3e7ee3e2006-04-02 23:11:04 +0200740 int ret;
Jens Axboe5274f052006-03-30 15:15:30 +0200741
742 /*
Jens Axboe49d0b212006-04-10 09:04:41 +0200743 * make sure the data in this buffer is uptodate
Jens Axboe5274f052006-03-30 15:15:30 +0200744 */
Jens Axboecac36bb2007-06-14 13:10:48 +0200745 ret = buf->ops->confirm(pipe, buf);
Jens Axboef84d7512006-05-01 19:59:03 +0200746 if (unlikely(ret))
747 return ret;
Jens Axboe5274f052006-03-30 15:15:30 +0200748
Jens Axboe5274f052006-03-30 15:15:30 +0200749 offset = sd->pos & ~PAGE_CACHE_MASK;
750
Jens Axboe016b6612006-04-25 15:42:00 +0200751 this_len = sd->len;
752 if (this_len + offset > PAGE_CACHE_SIZE)
753 this_len = PAGE_CACHE_SIZE - offset;
754
Nick Pigginafddba42007-10-16 01:25:01 -0700755 ret = pagecache_write_begin(file, mapping, sd->pos, this_len,
756 AOP_FLAG_UNINTERRUPTIBLE, &page, &fsdata);
757 if (unlikely(ret))
758 goto out;
Jens Axboe5274f052006-03-30 15:15:30 +0200759
Jens Axboe0568b402006-05-01 19:50:48 +0200760 if (buf->page != page) {
Jens Axboef84d7512006-05-01 19:59:03 +0200761 /*
762 * Careful, ->map() uses KM_USER0!
763 */
Jens Axboe76ad4d12006-05-03 10:41:33 +0200764 char *src = buf->ops->map(pipe, buf, 1);
Jens Axboef84d7512006-05-01 19:59:03 +0200765 char *dst = kmap_atomic(page, KM_USER1);
Jens Axboe5abc97a2006-03-30 15:16:46 +0200766
Jens Axboe016b6612006-04-25 15:42:00 +0200767 memcpy(dst + offset, src + buf->offset, this_len);
Jens Axboe5abc97a2006-03-30 15:16:46 +0200768 flush_dcache_page(page);
Jens Axboef84d7512006-05-01 19:59:03 +0200769 kunmap_atomic(dst, KM_USER1);
Jens Axboe76ad4d12006-05-03 10:41:33 +0200770 buf->ops->unmap(pipe, buf, src);
Jens Axboe5abc97a2006-03-30 15:16:46 +0200771 }
Nick Pigginafddba42007-10-16 01:25:01 -0700772 ret = pagecache_write_end(file, mapping, sd->pos, this_len, this_len,
773 page, fsdata);
Jens Axboe5274f052006-03-30 15:15:30 +0200774out:
Jens Axboe5274f052006-03-30 15:15:30 +0200775 return ret;
776}
Miklos Szeredi328eaab2009-04-14 19:48:39 +0200777EXPORT_SYMBOL(pipe_to_file);
Jens Axboe5274f052006-03-30 15:15:30 +0200778
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200779static void wakeup_pipe_writers(struct pipe_inode_info *pipe)
780{
781 smp_mb();
782 if (waitqueue_active(&pipe->wait))
783 wake_up_interruptible(&pipe->wait);
784 kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
785}
786
787/**
788 * splice_from_pipe_feed - feed available data from a pipe to a file
789 * @pipe: pipe to splice from
790 * @sd: information to @actor
791 * @actor: handler that splices the data
792 *
793 * Description:
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200794 * This function loops over the pipe and calls @actor to do the
795 * actual moving of a single struct pipe_buffer to the desired
796 * destination. It returns when there's no more buffers left in
797 * the pipe or if the requested number of bytes (@sd->total_len)
798 * have been copied. It returns a positive number (one) if the
799 * pipe needs to be filled with more data, zero if the required
800 * number of bytes have been copied and -errno on error.
801 *
802 * This, together with splice_from_pipe_{begin,end,next}, may be
803 * used to implement the functionality of __splice_from_pipe() when
804 * locking is required around copying the pipe buffers to the
805 * destination.
806 */
807int splice_from_pipe_feed(struct pipe_inode_info *pipe, struct splice_desc *sd,
808 splice_actor *actor)
809{
810 int ret;
811
812 while (pipe->nrbufs) {
813 struct pipe_buffer *buf = pipe->bufs + pipe->curbuf;
814 const struct pipe_buf_operations *ops = buf->ops;
815
816 sd->len = buf->len;
817 if (sd->len > sd->total_len)
818 sd->len = sd->total_len;
819
820 ret = actor(pipe, buf, sd);
821 if (ret <= 0) {
822 if (ret == -ENODATA)
823 ret = 0;
824 return ret;
825 }
826 buf->offset += ret;
827 buf->len -= ret;
828
829 sd->num_spliced += ret;
830 sd->len -= ret;
831 sd->pos += ret;
832 sd->total_len -= ret;
833
834 if (!buf->len) {
835 buf->ops = NULL;
836 ops->release(pipe, buf);
Jens Axboe35f3d142010-05-20 10:43:18 +0200837 pipe->curbuf = (pipe->curbuf + 1) & (pipe->buffers - 1);
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200838 pipe->nrbufs--;
839 if (pipe->inode)
840 sd->need_wakeup = true;
841 }
842
843 if (!sd->total_len)
844 return 0;
845 }
846
847 return 1;
848}
849EXPORT_SYMBOL(splice_from_pipe_feed);
850
851/**
852 * splice_from_pipe_next - wait for some data to splice from
853 * @pipe: pipe to splice from
854 * @sd: information about the splice operation
855 *
856 * Description:
857 * This function will wait for some data and return a positive
858 * value (one) if pipe buffers are available. It will return zero
859 * or -errno if no more data needs to be spliced.
860 */
861int splice_from_pipe_next(struct pipe_inode_info *pipe, struct splice_desc *sd)
862{
863 while (!pipe->nrbufs) {
864 if (!pipe->writers)
865 return 0;
866
867 if (!pipe->waiting_writers && sd->num_spliced)
868 return 0;
869
870 if (sd->flags & SPLICE_F_NONBLOCK)
871 return -EAGAIN;
872
873 if (signal_pending(current))
874 return -ERESTARTSYS;
875
876 if (sd->need_wakeup) {
877 wakeup_pipe_writers(pipe);
878 sd->need_wakeup = false;
879 }
880
881 pipe_wait(pipe);
882 }
883
884 return 1;
885}
886EXPORT_SYMBOL(splice_from_pipe_next);
887
888/**
889 * splice_from_pipe_begin - start splicing from pipe
Randy Dunlapb80901b2009-04-16 19:09:55 -0700890 * @sd: information about the splice operation
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200891 *
892 * Description:
893 * This function should be called before a loop containing
894 * splice_from_pipe_next() and splice_from_pipe_feed() to
895 * initialize the necessary fields of @sd.
896 */
897void splice_from_pipe_begin(struct splice_desc *sd)
898{
899 sd->num_spliced = 0;
900 sd->need_wakeup = false;
901}
902EXPORT_SYMBOL(splice_from_pipe_begin);
903
904/**
905 * splice_from_pipe_end - finish splicing from pipe
906 * @pipe: pipe to splice from
907 * @sd: information about the splice operation
908 *
909 * Description:
910 * This function will wake up pipe writers if necessary. It should
911 * be called after a loop containing splice_from_pipe_next() and
912 * splice_from_pipe_feed().
913 */
914void splice_from_pipe_end(struct pipe_inode_info *pipe, struct splice_desc *sd)
915{
916 if (sd->need_wakeup)
917 wakeup_pipe_writers(pipe);
918}
919EXPORT_SYMBOL(splice_from_pipe_end);
920
Jens Axboe932cc6d2007-06-21 13:10:21 +0200921/**
922 * __splice_from_pipe - splice data from a pipe to given actor
923 * @pipe: pipe to splice from
924 * @sd: information to @actor
925 * @actor: handler that splices the data
926 *
927 * Description:
928 * This function does little more than loop over the pipe and call
929 * @actor to do the actual moving of a single struct pipe_buffer to
930 * the desired destination. See pipe_to_file, pipe_to_sendpage, or
931 * pipe_to_user.
932 *
Jens Axboe83f91352006-04-02 23:05:09 +0200933 */
Jens Axboec66ab6f2007-06-12 21:17:17 +0200934ssize_t __splice_from_pipe(struct pipe_inode_info *pipe, struct splice_desc *sd,
935 splice_actor *actor)
Jens Axboe5274f052006-03-30 15:15:30 +0200936{
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200937 int ret;
Jens Axboe5274f052006-03-30 15:15:30 +0200938
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200939 splice_from_pipe_begin(sd);
940 do {
941 ret = splice_from_pipe_next(pipe, sd);
942 if (ret > 0)
943 ret = splice_from_pipe_feed(pipe, sd, actor);
944 } while (ret > 0);
945 splice_from_pipe_end(pipe, sd);
Jens Axboe5274f052006-03-30 15:15:30 +0200946
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200947 return sd->num_spliced ? sd->num_spliced : ret;
Jens Axboe5274f052006-03-30 15:15:30 +0200948}
Mark Fasheh40bee442007-03-21 13:11:02 +0100949EXPORT_SYMBOL(__splice_from_pipe);
Jens Axboe5274f052006-03-30 15:15:30 +0200950
Jens Axboe932cc6d2007-06-21 13:10:21 +0200951/**
952 * splice_from_pipe - splice data from a pipe to a file
953 * @pipe: pipe to splice from
954 * @out: file to splice to
955 * @ppos: position in @out
956 * @len: how many bytes to splice
957 * @flags: splice modifier flags
958 * @actor: handler that splices the data
959 *
960 * Description:
Miklos Szeredi29339702009-04-14 19:48:37 +0200961 * See __splice_from_pipe. This function locks the pipe inode,
Jens Axboe932cc6d2007-06-21 13:10:21 +0200962 * otherwise it's identical to __splice_from_pipe().
963 *
964 */
Mark Fasheh6da61802006-10-17 18:43:07 +0200965ssize_t splice_from_pipe(struct pipe_inode_info *pipe, struct file *out,
966 loff_t *ppos, size_t len, unsigned int flags,
967 splice_actor *actor)
968{
969 ssize_t ret;
Jens Axboec66ab6f2007-06-12 21:17:17 +0200970 struct splice_desc sd = {
971 .total_len = len,
972 .flags = flags,
973 .pos = *ppos,
Jens Axboe6a14b902007-06-14 13:08:55 +0200974 .u.file = out,
Jens Axboec66ab6f2007-06-12 21:17:17 +0200975 };
Mark Fasheh6da61802006-10-17 18:43:07 +0200976
Miklos Szeredi61e0d472009-04-14 19:48:41 +0200977 pipe_lock(pipe);
Jens Axboec66ab6f2007-06-12 21:17:17 +0200978 ret = __splice_from_pipe(pipe, &sd, actor);
Miklos Szeredi61e0d472009-04-14 19:48:41 +0200979 pipe_unlock(pipe);
Mark Fasheh6da61802006-10-17 18:43:07 +0200980
981 return ret;
982}
983
984/**
Jens Axboe83f91352006-04-02 23:05:09 +0200985 * generic_file_splice_write - splice data from a pipe to a file
Ingo Molnar3a326a22006-04-10 15:18:35 +0200986 * @pipe: pipe info
Jens Axboe83f91352006-04-02 23:05:09 +0200987 * @out: file to write to
Jens Axboe932cc6d2007-06-21 13:10:21 +0200988 * @ppos: position in @out
Jens Axboe83f91352006-04-02 23:05:09 +0200989 * @len: number of bytes to splice
990 * @flags: splice modifier flags
991 *
Jens Axboe932cc6d2007-06-21 13:10:21 +0200992 * Description:
993 * Will either move or copy pages (determined by @flags options) from
994 * the given pipe inode to the given file.
Jens Axboe83f91352006-04-02 23:05:09 +0200995 *
996 */
Ingo Molnar3a326a22006-04-10 15:18:35 +0200997ssize_t
998generic_file_splice_write(struct pipe_inode_info *pipe, struct file *out,
Jens Axboecbb7e572006-04-11 14:57:50 +0200999 loff_t *ppos, size_t len, unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +02001000{
Jens Axboe4f6f0bd2006-04-02 23:04:46 +02001001 struct address_space *mapping = out->f_mapping;
Jens Axboe8c34e2d2006-10-17 19:43:22 +02001002 struct inode *inode = mapping->host;
Miklos Szeredi7f3d4ee2008-05-07 09:22:39 +02001003 struct splice_desc sd = {
1004 .total_len = len,
1005 .flags = flags,
1006 .pos = *ppos,
1007 .u.file = out,
1008 };
Ingo Molnar3a326a22006-04-10 15:18:35 +02001009 ssize_t ret;
Jens Axboe8c34e2d2006-10-17 19:43:22 +02001010
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001011 pipe_lock(pipe);
Miklos Szeredieb443e52009-04-14 19:48:38 +02001012
1013 splice_from_pipe_begin(&sd);
1014 do {
1015 ret = splice_from_pipe_next(pipe, &sd);
1016 if (ret <= 0)
1017 break;
1018
1019 mutex_lock_nested(&inode->i_mutex, I_MUTEX_CHILD);
1020 ret = file_remove_suid(out);
Miklos Szeredi723590e2009-08-15 08:43:22 +02001021 if (!ret) {
1022 file_update_time(out);
Miklos Szeredieb443e52009-04-14 19:48:38 +02001023 ret = splice_from_pipe_feed(pipe, &sd, pipe_to_file);
Miklos Szeredi723590e2009-08-15 08:43:22 +02001024 }
Miklos Szeredieb443e52009-04-14 19:48:38 +02001025 mutex_unlock(&inode->i_mutex);
1026 } while (ret > 0);
1027 splice_from_pipe_end(pipe, &sd);
1028
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001029 pipe_unlock(pipe);
Miklos Szeredieb443e52009-04-14 19:48:38 +02001030
1031 if (sd.num_spliced)
1032 ret = sd.num_spliced;
1033
Jens Axboea4514eb2006-04-19 15:57:05 +02001034 if (ret > 0) {
Jens Axboe17ee4f42007-06-15 13:10:37 +02001035 unsigned long nr_pages;
Jan Kara148f9482009-08-17 19:52:36 +02001036 int err;
Jens Axboe17ee4f42007-06-15 13:10:37 +02001037
Jens Axboe17ee4f42007-06-15 13:10:37 +02001038 nr_pages = (ret + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
Jens Axboe4f6f0bd2006-04-02 23:04:46 +02001039
Jan Kara148f9482009-08-17 19:52:36 +02001040 err = generic_write_sync(out, *ppos, ret);
1041 if (err)
1042 ret = err;
1043 else
1044 *ppos += ret;
Jens Axboe17ee4f42007-06-15 13:10:37 +02001045 balance_dirty_pages_ratelimited_nr(mapping, nr_pages);
Jens Axboe4f6f0bd2006-04-02 23:04:46 +02001046 }
1047
1048 return ret;
Jens Axboe5274f052006-03-30 15:15:30 +02001049}
1050
Jens Axboe059a8f32006-04-02 23:06:05 +02001051EXPORT_SYMBOL(generic_file_splice_write);
1052
Miklos Szeredib2858d72009-05-19 11:37:46 +02001053static int write_pipe_buf(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
1054 struct splice_desc *sd)
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +02001055{
Miklos Szeredib2858d72009-05-19 11:37:46 +02001056 int ret;
1057 void *data;
1058
1059 ret = buf->ops->confirm(pipe, buf);
1060 if (ret)
1061 return ret;
1062
1063 data = buf->ops->map(pipe, buf, 0);
1064 ret = kernel_write(sd->u.file, data + buf->offset, sd->len, sd->pos);
1065 buf->ops->unmap(pipe, buf, data);
1066
1067 return ret;
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +02001068}
1069
1070static ssize_t default_file_splice_write(struct pipe_inode_info *pipe,
1071 struct file *out, loff_t *ppos,
1072 size_t len, unsigned int flags)
1073{
Miklos Szeredib2858d72009-05-19 11:37:46 +02001074 ssize_t ret;
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +02001075
Miklos Szeredib2858d72009-05-19 11:37:46 +02001076 ret = splice_from_pipe(pipe, out, ppos, len, flags, write_pipe_buf);
1077 if (ret > 0)
1078 *ppos += ret;
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +02001079
Miklos Szeredib2858d72009-05-19 11:37:46 +02001080 return ret;
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +02001081}
1082
Jens Axboe83f91352006-04-02 23:05:09 +02001083/**
1084 * generic_splice_sendpage - splice data from a pipe to a socket
Jens Axboe932cc6d2007-06-21 13:10:21 +02001085 * @pipe: pipe to splice from
Jens Axboe83f91352006-04-02 23:05:09 +02001086 * @out: socket to write to
Jens Axboe932cc6d2007-06-21 13:10:21 +02001087 * @ppos: position in @out
Jens Axboe83f91352006-04-02 23:05:09 +02001088 * @len: number of bytes to splice
1089 * @flags: splice modifier flags
1090 *
Jens Axboe932cc6d2007-06-21 13:10:21 +02001091 * Description:
1092 * Will send @len bytes from the pipe to a network socket. No data copying
1093 * is involved.
Jens Axboe83f91352006-04-02 23:05:09 +02001094 *
1095 */
Ingo Molnar3a326a22006-04-10 15:18:35 +02001096ssize_t generic_splice_sendpage(struct pipe_inode_info *pipe, struct file *out,
Jens Axboecbb7e572006-04-11 14:57:50 +02001097 loff_t *ppos, size_t len, unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +02001098{
Jens Axboe00522fb2006-04-26 14:39:29 +02001099 return splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_sendpage);
Jens Axboe5274f052006-03-30 15:15:30 +02001100}
1101
Jens Axboe059a8f32006-04-02 23:06:05 +02001102EXPORT_SYMBOL(generic_splice_sendpage);
Jeff Garzika0f06782006-03-30 23:06:13 -05001103
Jens Axboe83f91352006-04-02 23:05:09 +02001104/*
1105 * Attempt to initiate a splice from pipe to file.
1106 */
Ingo Molnar3a326a22006-04-10 15:18:35 +02001107static long do_splice_from(struct pipe_inode_info *pipe, struct file *out,
Jens Axboecbb7e572006-04-11 14:57:50 +02001108 loff_t *ppos, size_t len, unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +02001109{
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +02001110 ssize_t (*splice_write)(struct pipe_inode_info *, struct file *,
1111 loff_t *, size_t, unsigned int);
Jens Axboe5274f052006-03-30 15:15:30 +02001112 int ret;
1113
Jens Axboe49570e92006-04-11 13:56:09 +02001114 if (unlikely(!(out->f_mode & FMODE_WRITE)))
Jens Axboe5274f052006-03-30 15:15:30 +02001115 return -EBADF;
1116
Linus Torvaldsefc968d2008-10-09 14:04:54 -07001117 if (unlikely(out->f_flags & O_APPEND))
1118 return -EINVAL;
1119
Jens Axboecbb7e572006-04-11 14:57:50 +02001120 ret = rw_verify_area(WRITE, out, ppos, len);
Jens Axboe5274f052006-03-30 15:15:30 +02001121 if (unlikely(ret < 0))
1122 return ret;
1123
Changli Gaocc56f7d2009-11-04 09:09:52 +01001124 if (out->f_op && out->f_op->splice_write)
1125 splice_write = out->f_op->splice_write;
1126 else
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +02001127 splice_write = default_file_splice_write;
1128
1129 return splice_write(pipe, out, ppos, len, flags);
Jens Axboe5274f052006-03-30 15:15:30 +02001130}
1131
Jens Axboe83f91352006-04-02 23:05:09 +02001132/*
1133 * Attempt to initiate a splice from a file to a pipe.
1134 */
Jens Axboecbb7e572006-04-11 14:57:50 +02001135static long do_splice_to(struct file *in, loff_t *ppos,
1136 struct pipe_inode_info *pipe, size_t len,
1137 unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +02001138{
Miklos Szeredi68181732009-05-07 15:37:36 +02001139 ssize_t (*splice_read)(struct file *, loff_t *,
1140 struct pipe_inode_info *, size_t, unsigned int);
Jens Axboe5274f052006-03-30 15:15:30 +02001141 int ret;
1142
Jens Axboe49570e92006-04-11 13:56:09 +02001143 if (unlikely(!(in->f_mode & FMODE_READ)))
Jens Axboe5274f052006-03-30 15:15:30 +02001144 return -EBADF;
1145
Jens Axboecbb7e572006-04-11 14:57:50 +02001146 ret = rw_verify_area(READ, in, ppos, len);
Jens Axboe5274f052006-03-30 15:15:30 +02001147 if (unlikely(ret < 0))
1148 return ret;
1149
Changli Gaocc56f7d2009-11-04 09:09:52 +01001150 if (in->f_op && in->f_op->splice_read)
1151 splice_read = in->f_op->splice_read;
1152 else
Miklos Szeredi68181732009-05-07 15:37:36 +02001153 splice_read = default_file_splice_read;
1154
1155 return splice_read(in, ppos, pipe, len, flags);
Jens Axboe5274f052006-03-30 15:15:30 +02001156}
1157
Jens Axboe932cc6d2007-06-21 13:10:21 +02001158/**
1159 * splice_direct_to_actor - splices data directly between two non-pipes
1160 * @in: file to splice from
1161 * @sd: actor information on where to splice to
1162 * @actor: handles the data splicing
1163 *
1164 * Description:
1165 * This is a special case helper to splice directly between two
1166 * points, without requiring an explicit pipe. Internally an allocated
Randy Dunlap79685b82007-07-27 08:08:51 +02001167 * pipe is cached in the process, and reused during the lifetime of
Jens Axboe932cc6d2007-06-21 13:10:21 +02001168 * that process.
1169 *
Jens Axboec66ab6f2007-06-12 21:17:17 +02001170 */
1171ssize_t splice_direct_to_actor(struct file *in, struct splice_desc *sd,
1172 splice_direct_actor *actor)
Jens Axboeb92ce552006-04-11 13:52:07 +02001173{
1174 struct pipe_inode_info *pipe;
1175 long ret, bytes;
1176 umode_t i_mode;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001177 size_t len;
1178 int i, flags;
Jens Axboeb92ce552006-04-11 13:52:07 +02001179
1180 /*
1181 * We require the input being a regular file, as we don't want to
1182 * randomly drop data for eg socket -> socket splicing. Use the
1183 * piped splicing for that!
1184 */
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001185 i_mode = in->f_path.dentry->d_inode->i_mode;
Jens Axboeb92ce552006-04-11 13:52:07 +02001186 if (unlikely(!S_ISREG(i_mode) && !S_ISBLK(i_mode)))
1187 return -EINVAL;
1188
1189 /*
1190 * neither in nor out is a pipe, setup an internal pipe attached to
1191 * 'out' and transfer the wanted data from 'in' to 'out' through that
1192 */
1193 pipe = current->splice_pipe;
Jens Axboe49570e92006-04-11 13:56:09 +02001194 if (unlikely(!pipe)) {
Jens Axboeb92ce552006-04-11 13:52:07 +02001195 pipe = alloc_pipe_info(NULL);
1196 if (!pipe)
1197 return -ENOMEM;
1198
1199 /*
1200 * We don't have an immediate reader, but we'll read the stuff
Jens Axboe00522fb2006-04-26 14:39:29 +02001201 * out of the pipe right after the splice_to_pipe(). So set
Jens Axboeb92ce552006-04-11 13:52:07 +02001202 * PIPE_READERS appropriately.
1203 */
1204 pipe->readers = 1;
1205
1206 current->splice_pipe = pipe;
1207 }
1208
1209 /*
Ingo Molnar73d62d82006-04-11 13:57:21 +02001210 * Do the splice.
Jens Axboeb92ce552006-04-11 13:52:07 +02001211 */
1212 ret = 0;
1213 bytes = 0;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001214 len = sd->total_len;
1215 flags = sd->flags;
1216
1217 /*
1218 * Don't block on output, we have to drain the direct pipe.
1219 */
1220 sd->flags &= ~SPLICE_F_NONBLOCK;
Jens Axboeb92ce552006-04-11 13:52:07 +02001221
1222 while (len) {
Jens Axboe51a92c02007-07-13 14:11:43 +02001223 size_t read_len;
Tom Zanussia82c53a2008-05-09 13:28:36 +02001224 loff_t pos = sd->pos, prev_pos = pos;
Jens Axboeb92ce552006-04-11 13:52:07 +02001225
Jens Axboebcd4f3a2007-07-16 14:41:49 +02001226 ret = do_splice_to(in, &pos, pipe, len, flags);
Jens Axboe51a92c02007-07-13 14:11:43 +02001227 if (unlikely(ret <= 0))
Jens Axboeb92ce552006-04-11 13:52:07 +02001228 goto out_release;
1229
1230 read_len = ret;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001231 sd->total_len = read_len;
Jens Axboeb92ce552006-04-11 13:52:07 +02001232
1233 /*
1234 * NOTE: nonblocking mode only applies to the input. We
1235 * must not do the output in nonblocking mode as then we
1236 * could get stuck data in the internal pipe:
1237 */
Jens Axboec66ab6f2007-06-12 21:17:17 +02001238 ret = actor(pipe, sd);
Tom Zanussia82c53a2008-05-09 13:28:36 +02001239 if (unlikely(ret <= 0)) {
1240 sd->pos = prev_pos;
Jens Axboeb92ce552006-04-11 13:52:07 +02001241 goto out_release;
Tom Zanussia82c53a2008-05-09 13:28:36 +02001242 }
Jens Axboeb92ce552006-04-11 13:52:07 +02001243
1244 bytes += ret;
1245 len -= ret;
Jens Axboebcd4f3a2007-07-16 14:41:49 +02001246 sd->pos = pos;
Jens Axboeb92ce552006-04-11 13:52:07 +02001247
Tom Zanussia82c53a2008-05-09 13:28:36 +02001248 if (ret < read_len) {
1249 sd->pos = prev_pos + ret;
Jens Axboe51a92c02007-07-13 14:11:43 +02001250 goto out_release;
Tom Zanussia82c53a2008-05-09 13:28:36 +02001251 }
Jens Axboeb92ce552006-04-11 13:52:07 +02001252 }
1253
Jens Axboe9e971982008-01-29 21:05:57 +01001254done:
Jens Axboeb92ce552006-04-11 13:52:07 +02001255 pipe->nrbufs = pipe->curbuf = 0;
Jens Axboe80848702008-01-30 12:24:48 +01001256 file_accessed(in);
Jens Axboeb92ce552006-04-11 13:52:07 +02001257 return bytes;
1258
1259out_release:
1260 /*
1261 * If we did an incomplete transfer we must release
1262 * the pipe buffers in question:
1263 */
Jens Axboe35f3d142010-05-20 10:43:18 +02001264 for (i = 0; i < pipe->buffers; i++) {
Jens Axboeb92ce552006-04-11 13:52:07 +02001265 struct pipe_buffer *buf = pipe->bufs + i;
1266
1267 if (buf->ops) {
1268 buf->ops->release(pipe, buf);
1269 buf->ops = NULL;
1270 }
1271 }
Jens Axboeb92ce552006-04-11 13:52:07 +02001272
Jens Axboe9e971982008-01-29 21:05:57 +01001273 if (!bytes)
1274 bytes = ret;
Jens Axboeb92ce552006-04-11 13:52:07 +02001275
Jens Axboe9e971982008-01-29 21:05:57 +01001276 goto done;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001277}
1278EXPORT_SYMBOL(splice_direct_to_actor);
1279
1280static int direct_splice_actor(struct pipe_inode_info *pipe,
1281 struct splice_desc *sd)
1282{
Jens Axboe6a14b902007-06-14 13:08:55 +02001283 struct file *file = sd->u.file;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001284
1285 return do_splice_from(pipe, file, &sd->pos, sd->total_len, sd->flags);
1286}
1287
Jens Axboe932cc6d2007-06-21 13:10:21 +02001288/**
1289 * do_splice_direct - splices data directly between two files
1290 * @in: file to splice from
1291 * @ppos: input file offset
1292 * @out: file to splice to
1293 * @len: number of bytes to splice
1294 * @flags: splice modifier flags
1295 *
1296 * Description:
1297 * For use by do_sendfile(). splice can easily emulate sendfile, but
1298 * doing it in the application would incur an extra system call
1299 * (splice in + splice out, as compared to just sendfile()). So this helper
1300 * can splice directly through a process-private pipe.
1301 *
1302 */
Jens Axboec66ab6f2007-06-12 21:17:17 +02001303long do_splice_direct(struct file *in, loff_t *ppos, struct file *out,
1304 size_t len, unsigned int flags)
1305{
1306 struct splice_desc sd = {
1307 .len = len,
1308 .total_len = len,
1309 .flags = flags,
1310 .pos = *ppos,
Jens Axboe6a14b902007-06-14 13:08:55 +02001311 .u.file = out,
Jens Axboec66ab6f2007-06-12 21:17:17 +02001312 };
Jens Axboe51a92c02007-07-13 14:11:43 +02001313 long ret;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001314
1315 ret = splice_direct_to_actor(in, &sd, direct_splice_actor);
Jens Axboe51a92c02007-07-13 14:11:43 +02001316 if (ret > 0)
Tom Zanussia82c53a2008-05-09 13:28:36 +02001317 *ppos = sd.pos;
Jens Axboe51a92c02007-07-13 14:11:43 +02001318
Jens Axboec66ab6f2007-06-12 21:17:17 +02001319 return ret;
Jens Axboeb92ce552006-04-11 13:52:07 +02001320}
1321
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001322static int splice_pipe_to_pipe(struct pipe_inode_info *ipipe,
1323 struct pipe_inode_info *opipe,
1324 size_t len, unsigned int flags);
Jens Axboe83f91352006-04-02 23:05:09 +02001325/*
Jens Axboeddac0d32006-11-04 12:49:32 +01001326 * After the inode slimming patch, i_pipe/i_bdev/i_cdev share the same
1327 * location, so checking ->i_pipe is not enough to verify that this is a
1328 * pipe.
1329 */
1330static inline struct pipe_inode_info *pipe_info(struct inode *inode)
1331{
1332 if (S_ISFIFO(inode->i_mode))
1333 return inode->i_pipe;
1334
1335 return NULL;
1336}
1337
1338/*
Jens Axboe83f91352006-04-02 23:05:09 +02001339 * Determine where to splice to/from.
1340 */
Ingo Molnar529565d2006-04-10 15:18:58 +02001341static long do_splice(struct file *in, loff_t __user *off_in,
1342 struct file *out, loff_t __user *off_out,
1343 size_t len, unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +02001344{
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001345 struct pipe_inode_info *ipipe;
1346 struct pipe_inode_info *opipe;
Jens Axboecbb7e572006-04-11 14:57:50 +02001347 loff_t offset, *off;
Jens Axboea4514eb2006-04-19 15:57:05 +02001348 long ret;
Jens Axboe5274f052006-03-30 15:15:30 +02001349
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001350 ipipe = pipe_info(in->f_path.dentry->d_inode);
1351 opipe = pipe_info(out->f_path.dentry->d_inode);
1352
1353 if (ipipe && opipe) {
1354 if (off_in || off_out)
1355 return -ESPIPE;
1356
1357 if (!(in->f_mode & FMODE_READ))
1358 return -EBADF;
1359
1360 if (!(out->f_mode & FMODE_WRITE))
1361 return -EBADF;
1362
1363 /* Splicing to self would be fun, but... */
1364 if (ipipe == opipe)
1365 return -EINVAL;
1366
1367 return splice_pipe_to_pipe(ipipe, opipe, len, flags);
1368 }
1369
1370 if (ipipe) {
Ingo Molnar529565d2006-04-10 15:18:58 +02001371 if (off_in)
1372 return -ESPIPE;
Jens Axboeb92ce552006-04-11 13:52:07 +02001373 if (off_out) {
Changli Gaocc56f7d2009-11-04 09:09:52 +01001374 if (!out->f_op || !out->f_op->llseek ||
1375 out->f_op->llseek == no_llseek)
Jens Axboeb92ce552006-04-11 13:52:07 +02001376 return -EINVAL;
Jens Axboecbb7e572006-04-11 14:57:50 +02001377 if (copy_from_user(&offset, off_out, sizeof(loff_t)))
Jens Axboeb92ce552006-04-11 13:52:07 +02001378 return -EFAULT;
Jens Axboecbb7e572006-04-11 14:57:50 +02001379 off = &offset;
1380 } else
1381 off = &out->f_pos;
Ingo Molnar529565d2006-04-10 15:18:58 +02001382
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001383 ret = do_splice_from(ipipe, out, off, len, flags);
Jens Axboea4514eb2006-04-19 15:57:05 +02001384
1385 if (off_out && copy_to_user(off_out, off, sizeof(loff_t)))
1386 ret = -EFAULT;
1387
1388 return ret;
Ingo Molnar529565d2006-04-10 15:18:58 +02001389 }
Jens Axboe5274f052006-03-30 15:15:30 +02001390
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001391 if (opipe) {
Ingo Molnar529565d2006-04-10 15:18:58 +02001392 if (off_out)
1393 return -ESPIPE;
Jens Axboeb92ce552006-04-11 13:52:07 +02001394 if (off_in) {
Changli Gaocc56f7d2009-11-04 09:09:52 +01001395 if (!in->f_op || !in->f_op->llseek ||
1396 in->f_op->llseek == no_llseek)
Jens Axboeb92ce552006-04-11 13:52:07 +02001397 return -EINVAL;
Jens Axboecbb7e572006-04-11 14:57:50 +02001398 if (copy_from_user(&offset, off_in, sizeof(loff_t)))
Jens Axboeb92ce552006-04-11 13:52:07 +02001399 return -EFAULT;
Jens Axboecbb7e572006-04-11 14:57:50 +02001400 off = &offset;
1401 } else
1402 off = &in->f_pos;
Ingo Molnar529565d2006-04-10 15:18:58 +02001403
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001404 ret = do_splice_to(in, off, opipe, len, flags);
Jens Axboea4514eb2006-04-19 15:57:05 +02001405
1406 if (off_in && copy_to_user(off_in, off, sizeof(loff_t)))
1407 ret = -EFAULT;
1408
1409 return ret;
Ingo Molnar529565d2006-04-10 15:18:58 +02001410 }
Jens Axboe5274f052006-03-30 15:15:30 +02001411
1412 return -EINVAL;
1413}
1414
Jens Axboe912d35f2006-04-26 10:59:21 +02001415/*
1416 * Map an iov into an array of pages and offset/length tupples. With the
1417 * partial_page structure, we can map several non-contiguous ranges into
1418 * our ones pages[] map instead of splitting that operation into pieces.
1419 * Could easily be exported as a generic helper for other users, in which
1420 * case one would probably want to add a 'max_nr_pages' parameter as well.
1421 */
1422static int get_iovec_page_array(const struct iovec __user *iov,
1423 unsigned int nr_vecs, struct page **pages,
Jens Axboe35f3d142010-05-20 10:43:18 +02001424 struct partial_page *partial, int aligned,
1425 unsigned int pipe_buffers)
Jens Axboe912d35f2006-04-26 10:59:21 +02001426{
1427 int buffers = 0, error = 0;
1428
Jens Axboe912d35f2006-04-26 10:59:21 +02001429 while (nr_vecs) {
1430 unsigned long off, npages;
Linus Torvalds75723952007-10-01 13:17:28 -07001431 struct iovec entry;
Jens Axboe912d35f2006-04-26 10:59:21 +02001432 void __user *base;
1433 size_t len;
1434 int i;
1435
Linus Torvalds75723952007-10-01 13:17:28 -07001436 error = -EFAULT;
Nick Pigginbc40d732008-07-25 19:45:26 -07001437 if (copy_from_user(&entry, iov, sizeof(entry)))
Jens Axboe912d35f2006-04-26 10:59:21 +02001438 break;
Linus Torvalds75723952007-10-01 13:17:28 -07001439
1440 base = entry.iov_base;
1441 len = entry.iov_len;
Jens Axboe912d35f2006-04-26 10:59:21 +02001442
1443 /*
1444 * Sanity check this iovec. 0 read succeeds.
1445 */
Linus Torvalds75723952007-10-01 13:17:28 -07001446 error = 0;
Jens Axboe912d35f2006-04-26 10:59:21 +02001447 if (unlikely(!len))
1448 break;
1449 error = -EFAULT;
Bastian Blank712a30e2008-02-10 16:47:57 +02001450 if (!access_ok(VERIFY_READ, base, len))
Jens Axboe912d35f2006-04-26 10:59:21 +02001451 break;
1452
1453 /*
1454 * Get this base offset and number of pages, then map
1455 * in the user pages.
1456 */
1457 off = (unsigned long) base & ~PAGE_MASK;
Jens Axboe7afa6fd2006-05-01 20:02:33 +02001458
1459 /*
1460 * If asked for alignment, the offset must be zero and the
1461 * length a multiple of the PAGE_SIZE.
1462 */
1463 error = -EINVAL;
1464 if (aligned && (off || len & ~PAGE_MASK))
1465 break;
1466
Jens Axboe912d35f2006-04-26 10:59:21 +02001467 npages = (off + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
Jens Axboe35f3d142010-05-20 10:43:18 +02001468 if (npages > pipe_buffers - buffers)
1469 npages = pipe_buffers - buffers;
Jens Axboe912d35f2006-04-26 10:59:21 +02001470
Nick Pigginbc40d732008-07-25 19:45:26 -07001471 error = get_user_pages_fast((unsigned long)base, npages,
1472 0, &pages[buffers]);
Jens Axboe912d35f2006-04-26 10:59:21 +02001473
1474 if (unlikely(error <= 0))
1475 break;
1476
1477 /*
1478 * Fill this contiguous range into the partial page map.
1479 */
1480 for (i = 0; i < error; i++) {
Jens Axboe75914892006-05-02 12:57:18 +02001481 const int plen = min_t(size_t, len, PAGE_SIZE - off);
Jens Axboe912d35f2006-04-26 10:59:21 +02001482
1483 partial[buffers].offset = off;
1484 partial[buffers].len = plen;
1485
1486 off = 0;
1487 len -= plen;
1488 buffers++;
1489 }
1490
1491 /*
1492 * We didn't complete this iov, stop here since it probably
1493 * means we have to move some of this into a pipe to
1494 * be able to continue.
1495 */
1496 if (len)
1497 break;
1498
1499 /*
1500 * Don't continue if we mapped fewer pages than we asked for,
1501 * or if we mapped the max number of pages that we have
1502 * room for.
1503 */
Jens Axboe35f3d142010-05-20 10:43:18 +02001504 if (error < npages || buffers == pipe_buffers)
Jens Axboe912d35f2006-04-26 10:59:21 +02001505 break;
1506
1507 nr_vecs--;
1508 iov++;
1509 }
1510
Jens Axboe912d35f2006-04-26 10:59:21 +02001511 if (buffers)
1512 return buffers;
1513
1514 return error;
1515}
1516
Jens Axboe6a14b902007-06-14 13:08:55 +02001517static int pipe_to_user(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
1518 struct splice_desc *sd)
1519{
1520 char *src;
1521 int ret;
1522
Jens Axboecac36bb2007-06-14 13:10:48 +02001523 ret = buf->ops->confirm(pipe, buf);
Jens Axboe6a14b902007-06-14 13:08:55 +02001524 if (unlikely(ret))
1525 return ret;
1526
1527 /*
1528 * See if we can use the atomic maps, by prefaulting in the
1529 * pages and doing an atomic copy
1530 */
1531 if (!fault_in_pages_writeable(sd->u.userptr, sd->len)) {
1532 src = buf->ops->map(pipe, buf, 1);
1533 ret = __copy_to_user_inatomic(sd->u.userptr, src + buf->offset,
1534 sd->len);
1535 buf->ops->unmap(pipe, buf, src);
1536 if (!ret) {
1537 ret = sd->len;
1538 goto out;
1539 }
1540 }
1541
1542 /*
1543 * No dice, use slow non-atomic map and copy
1544 */
1545 src = buf->ops->map(pipe, buf, 0);
1546
1547 ret = sd->len;
1548 if (copy_to_user(sd->u.userptr, src + buf->offset, sd->len))
1549 ret = -EFAULT;
1550
Jens Axboe6866bef2007-10-16 10:01:29 +02001551 buf->ops->unmap(pipe, buf, src);
Jens Axboe6a14b902007-06-14 13:08:55 +02001552out:
1553 if (ret > 0)
1554 sd->u.userptr += ret;
Jens Axboe6a14b902007-06-14 13:08:55 +02001555 return ret;
1556}
1557
1558/*
1559 * For lack of a better implementation, implement vmsplice() to userspace
1560 * as a simple copy of the pipes pages to the user iov.
1561 */
1562static long vmsplice_to_user(struct file *file, const struct iovec __user *iov,
1563 unsigned long nr_segs, unsigned int flags)
1564{
1565 struct pipe_inode_info *pipe;
1566 struct splice_desc sd;
1567 ssize_t size;
1568 int error;
1569 long ret;
1570
1571 pipe = pipe_info(file->f_path.dentry->d_inode);
1572 if (!pipe)
1573 return -EBADF;
1574
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001575 pipe_lock(pipe);
Jens Axboe6a14b902007-06-14 13:08:55 +02001576
1577 error = ret = 0;
1578 while (nr_segs) {
1579 void __user *base;
1580 size_t len;
1581
1582 /*
1583 * Get user address base and length for this iovec.
1584 */
1585 error = get_user(base, &iov->iov_base);
1586 if (unlikely(error))
1587 break;
1588 error = get_user(len, &iov->iov_len);
1589 if (unlikely(error))
1590 break;
1591
1592 /*
1593 * Sanity check this iovec. 0 read succeeds.
1594 */
1595 if (unlikely(!len))
1596 break;
1597 if (unlikely(!base)) {
1598 error = -EFAULT;
1599 break;
1600 }
1601
Jens Axboe88119302008-02-08 08:49:14 -08001602 if (unlikely(!access_ok(VERIFY_WRITE, base, len))) {
1603 error = -EFAULT;
1604 break;
1605 }
1606
Jens Axboe6a14b902007-06-14 13:08:55 +02001607 sd.len = 0;
1608 sd.total_len = len;
1609 sd.flags = flags;
1610 sd.u.userptr = base;
1611 sd.pos = 0;
1612
1613 size = __splice_from_pipe(pipe, &sd, pipe_to_user);
1614 if (size < 0) {
1615 if (!ret)
1616 ret = size;
1617
1618 break;
1619 }
1620
1621 ret += size;
1622
1623 if (size < len)
1624 break;
1625
1626 nr_segs--;
1627 iov++;
1628 }
1629
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001630 pipe_unlock(pipe);
Jens Axboe6a14b902007-06-14 13:08:55 +02001631
1632 if (!ret)
1633 ret = error;
1634
1635 return ret;
1636}
1637
Jens Axboe912d35f2006-04-26 10:59:21 +02001638/*
1639 * vmsplice splices a user address range into a pipe. It can be thought of
1640 * as splice-from-memory, where the regular splice is splice-from-file (or
1641 * to file). In both cases the output is a pipe, naturally.
Jens Axboe912d35f2006-04-26 10:59:21 +02001642 */
Jens Axboe6a14b902007-06-14 13:08:55 +02001643static long vmsplice_to_pipe(struct file *file, const struct iovec __user *iov,
1644 unsigned long nr_segs, unsigned int flags)
Jens Axboe912d35f2006-04-26 10:59:21 +02001645{
Jens Axboeddac0d32006-11-04 12:49:32 +01001646 struct pipe_inode_info *pipe;
Jens Axboe35f3d142010-05-20 10:43:18 +02001647 struct page *pages[PIPE_DEF_BUFFERS];
1648 struct partial_page partial[PIPE_DEF_BUFFERS];
Jens Axboe912d35f2006-04-26 10:59:21 +02001649 struct splice_pipe_desc spd = {
1650 .pages = pages,
1651 .partial = partial,
1652 .flags = flags,
1653 .ops = &user_page_pipe_buf_ops,
Jens Axboebbdfc2f2007-11-06 23:29:47 -08001654 .spd_release = spd_release_page,
Jens Axboe912d35f2006-04-26 10:59:21 +02001655 };
Jens Axboe35f3d142010-05-20 10:43:18 +02001656 long ret;
Jens Axboe912d35f2006-04-26 10:59:21 +02001657
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001658 pipe = pipe_info(file->f_path.dentry->d_inode);
Jens Axboeddac0d32006-11-04 12:49:32 +01001659 if (!pipe)
Jens Axboe912d35f2006-04-26 10:59:21 +02001660 return -EBADF;
Jens Axboe912d35f2006-04-26 10:59:21 +02001661
Jens Axboe35f3d142010-05-20 10:43:18 +02001662 if (splice_grow_spd(pipe, &spd))
1663 return -ENOMEM;
Jens Axboe912d35f2006-04-26 10:59:21 +02001664
Jens Axboe35f3d142010-05-20 10:43:18 +02001665 spd.nr_pages = get_iovec_page_array(iov, nr_segs, spd.pages,
1666 spd.partial, flags & SPLICE_F_GIFT,
1667 pipe->buffers);
1668 if (spd.nr_pages <= 0)
1669 ret = spd.nr_pages;
1670 else
1671 ret = splice_to_pipe(pipe, &spd);
1672
1673 splice_shrink_spd(pipe, &spd);
1674 return ret;
Jens Axboe912d35f2006-04-26 10:59:21 +02001675}
1676
Jens Axboe6a14b902007-06-14 13:08:55 +02001677/*
1678 * Note that vmsplice only really supports true splicing _from_ user memory
1679 * to a pipe, not the other way around. Splicing from user memory is a simple
1680 * operation that can be supported without any funky alignment restrictions
1681 * or nasty vm tricks. We simply map in the user memory and fill them into
1682 * a pipe. The reverse isn't quite as easy, though. There are two possible
1683 * solutions for that:
1684 *
1685 * - memcpy() the data internally, at which point we might as well just
1686 * do a regular read() on the buffer anyway.
1687 * - Lots of nasty vm tricks, that are neither fast nor flexible (it
1688 * has restriction limitations on both ends of the pipe).
1689 *
1690 * Currently we punt and implement it as a normal copy, see pipe_to_user().
1691 *
1692 */
Heiko Carstens836f92a2009-01-14 14:14:33 +01001693SYSCALL_DEFINE4(vmsplice, int, fd, const struct iovec __user *, iov,
1694 unsigned long, nr_segs, unsigned int, flags)
Jens Axboe912d35f2006-04-26 10:59:21 +02001695{
1696 struct file *file;
1697 long error;
1698 int fput;
1699
Jens Axboe6a14b902007-06-14 13:08:55 +02001700 if (unlikely(nr_segs > UIO_MAXIOV))
1701 return -EINVAL;
1702 else if (unlikely(!nr_segs))
1703 return 0;
1704
Jens Axboe912d35f2006-04-26 10:59:21 +02001705 error = -EBADF;
1706 file = fget_light(fd, &fput);
1707 if (file) {
1708 if (file->f_mode & FMODE_WRITE)
Jens Axboe6a14b902007-06-14 13:08:55 +02001709 error = vmsplice_to_pipe(file, iov, nr_segs, flags);
1710 else if (file->f_mode & FMODE_READ)
1711 error = vmsplice_to_user(file, iov, nr_segs, flags);
Jens Axboe912d35f2006-04-26 10:59:21 +02001712
1713 fput_light(file, fput);
1714 }
1715
1716 return error;
1717}
1718
Heiko Carstens836f92a2009-01-14 14:14:33 +01001719SYSCALL_DEFINE6(splice, int, fd_in, loff_t __user *, off_in,
1720 int, fd_out, loff_t __user *, off_out,
1721 size_t, len, unsigned int, flags)
Jens Axboe5274f052006-03-30 15:15:30 +02001722{
1723 long error;
1724 struct file *in, *out;
1725 int fput_in, fput_out;
1726
1727 if (unlikely(!len))
1728 return 0;
1729
1730 error = -EBADF;
Ingo Molnar529565d2006-04-10 15:18:58 +02001731 in = fget_light(fd_in, &fput_in);
Jens Axboe5274f052006-03-30 15:15:30 +02001732 if (in) {
1733 if (in->f_mode & FMODE_READ) {
Ingo Molnar529565d2006-04-10 15:18:58 +02001734 out = fget_light(fd_out, &fput_out);
Jens Axboe5274f052006-03-30 15:15:30 +02001735 if (out) {
1736 if (out->f_mode & FMODE_WRITE)
Ingo Molnar529565d2006-04-10 15:18:58 +02001737 error = do_splice(in, off_in,
1738 out, off_out,
1739 len, flags);
Jens Axboe5274f052006-03-30 15:15:30 +02001740 fput_light(out, fput_out);
1741 }
1742 }
1743
1744 fput_light(in, fput_in);
1745 }
1746
1747 return error;
1748}
Jens Axboe70524492006-04-11 15:51:17 +02001749
1750/*
Jens Axboeaadd06e2006-07-10 11:00:01 +02001751 * Make sure there's data to read. Wait for input if we can, otherwise
1752 * return an appropriate error.
1753 */
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001754static int ipipe_prep(struct pipe_inode_info *pipe, unsigned int flags)
Jens Axboeaadd06e2006-07-10 11:00:01 +02001755{
1756 int ret;
1757
1758 /*
1759 * Check ->nrbufs without the inode lock first. This function
1760 * is speculative anyways, so missing one is ok.
1761 */
1762 if (pipe->nrbufs)
1763 return 0;
1764
1765 ret = 0;
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001766 pipe_lock(pipe);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001767
1768 while (!pipe->nrbufs) {
1769 if (signal_pending(current)) {
1770 ret = -ERESTARTSYS;
1771 break;
1772 }
1773 if (!pipe->writers)
1774 break;
1775 if (!pipe->waiting_writers) {
1776 if (flags & SPLICE_F_NONBLOCK) {
1777 ret = -EAGAIN;
1778 break;
1779 }
1780 }
1781 pipe_wait(pipe);
1782 }
1783
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001784 pipe_unlock(pipe);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001785 return ret;
1786}
1787
1788/*
1789 * Make sure there's writeable room. Wait for room if we can, otherwise
1790 * return an appropriate error.
1791 */
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001792static int opipe_prep(struct pipe_inode_info *pipe, unsigned int flags)
Jens Axboeaadd06e2006-07-10 11:00:01 +02001793{
1794 int ret;
1795
1796 /*
1797 * Check ->nrbufs without the inode lock first. This function
1798 * is speculative anyways, so missing one is ok.
1799 */
Jens Axboe35f3d142010-05-20 10:43:18 +02001800 if (pipe->nrbufs < pipe->buffers)
Jens Axboeaadd06e2006-07-10 11:00:01 +02001801 return 0;
1802
1803 ret = 0;
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001804 pipe_lock(pipe);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001805
Jens Axboe35f3d142010-05-20 10:43:18 +02001806 while (pipe->nrbufs >= pipe->buffers) {
Jens Axboeaadd06e2006-07-10 11:00:01 +02001807 if (!pipe->readers) {
1808 send_sig(SIGPIPE, current, 0);
1809 ret = -EPIPE;
1810 break;
1811 }
1812 if (flags & SPLICE_F_NONBLOCK) {
1813 ret = -EAGAIN;
1814 break;
1815 }
1816 if (signal_pending(current)) {
1817 ret = -ERESTARTSYS;
1818 break;
1819 }
1820 pipe->waiting_writers++;
1821 pipe_wait(pipe);
1822 pipe->waiting_writers--;
1823 }
1824
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001825 pipe_unlock(pipe);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001826 return ret;
1827}
1828
1829/*
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001830 * Splice contents of ipipe to opipe.
1831 */
1832static int splice_pipe_to_pipe(struct pipe_inode_info *ipipe,
1833 struct pipe_inode_info *opipe,
1834 size_t len, unsigned int flags)
1835{
1836 struct pipe_buffer *ibuf, *obuf;
1837 int ret = 0, nbuf;
1838 bool input_wakeup = false;
1839
1840
1841retry:
1842 ret = ipipe_prep(ipipe, flags);
1843 if (ret)
1844 return ret;
1845
1846 ret = opipe_prep(opipe, flags);
1847 if (ret)
1848 return ret;
1849
1850 /*
1851 * Potential ABBA deadlock, work around it by ordering lock
1852 * grabbing by pipe info address. Otherwise two different processes
1853 * could deadlock (one doing tee from A -> B, the other from B -> A).
1854 */
1855 pipe_double_lock(ipipe, opipe);
1856
1857 do {
1858 if (!opipe->readers) {
1859 send_sig(SIGPIPE, current, 0);
1860 if (!ret)
1861 ret = -EPIPE;
1862 break;
1863 }
1864
1865 if (!ipipe->nrbufs && !ipipe->writers)
1866 break;
1867
1868 /*
1869 * Cannot make any progress, because either the input
1870 * pipe is empty or the output pipe is full.
1871 */
Jens Axboe35f3d142010-05-20 10:43:18 +02001872 if (!ipipe->nrbufs || opipe->nrbufs >= opipe->buffers) {
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001873 /* Already processed some buffers, break */
1874 if (ret)
1875 break;
1876
1877 if (flags & SPLICE_F_NONBLOCK) {
1878 ret = -EAGAIN;
1879 break;
1880 }
1881
1882 /*
1883 * We raced with another reader/writer and haven't
1884 * managed to process any buffers. A zero return
1885 * value means EOF, so retry instead.
1886 */
1887 pipe_unlock(ipipe);
1888 pipe_unlock(opipe);
1889 goto retry;
1890 }
1891
1892 ibuf = ipipe->bufs + ipipe->curbuf;
Jens Axboe35f3d142010-05-20 10:43:18 +02001893 nbuf = (opipe->curbuf + opipe->nrbufs) & (opipe->buffers - 1);
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001894 obuf = opipe->bufs + nbuf;
1895
1896 if (len >= ibuf->len) {
1897 /*
1898 * Simply move the whole buffer from ipipe to opipe
1899 */
1900 *obuf = *ibuf;
1901 ibuf->ops = NULL;
1902 opipe->nrbufs++;
Jens Axboe35f3d142010-05-20 10:43:18 +02001903 ipipe->curbuf = (ipipe->curbuf + 1) & (ipipe->buffers - 1);
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001904 ipipe->nrbufs--;
1905 input_wakeup = true;
1906 } else {
1907 /*
1908 * Get a reference to this pipe buffer,
1909 * so we can copy the contents over.
1910 */
1911 ibuf->ops->get(ipipe, ibuf);
1912 *obuf = *ibuf;
1913
1914 /*
1915 * Don't inherit the gift flag, we need to
1916 * prevent multiple steals of this page.
1917 */
1918 obuf->flags &= ~PIPE_BUF_FLAG_GIFT;
1919
1920 obuf->len = len;
1921 opipe->nrbufs++;
1922 ibuf->offset += obuf->len;
1923 ibuf->len -= obuf->len;
1924 }
1925 ret += obuf->len;
1926 len -= obuf->len;
1927 } while (len);
1928
1929 pipe_unlock(ipipe);
1930 pipe_unlock(opipe);
1931
1932 /*
1933 * If we put data in the output pipe, wakeup any potential readers.
1934 */
1935 if (ret > 0) {
1936 smp_mb();
1937 if (waitqueue_active(&opipe->wait))
1938 wake_up_interruptible(&opipe->wait);
1939 kill_fasync(&opipe->fasync_readers, SIGIO, POLL_IN);
1940 }
1941 if (input_wakeup)
1942 wakeup_pipe_writers(ipipe);
1943
1944 return ret;
1945}
1946
1947/*
Jens Axboe70524492006-04-11 15:51:17 +02001948 * Link contents of ipipe to opipe.
1949 */
1950static int link_pipe(struct pipe_inode_info *ipipe,
1951 struct pipe_inode_info *opipe,
1952 size_t len, unsigned int flags)
1953{
1954 struct pipe_buffer *ibuf, *obuf;
Jens Axboeaadd06e2006-07-10 11:00:01 +02001955 int ret = 0, i = 0, nbuf;
Jens Axboe70524492006-04-11 15:51:17 +02001956
1957 /*
1958 * Potential ABBA deadlock, work around it by ordering lock
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001959 * grabbing by pipe info address. Otherwise two different processes
Jens Axboe70524492006-04-11 15:51:17 +02001960 * could deadlock (one doing tee from A -> B, the other from B -> A).
1961 */
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001962 pipe_double_lock(ipipe, opipe);
Jens Axboe70524492006-04-11 15:51:17 +02001963
Jens Axboeaadd06e2006-07-10 11:00:01 +02001964 do {
Jens Axboe70524492006-04-11 15:51:17 +02001965 if (!opipe->readers) {
1966 send_sig(SIGPIPE, current, 0);
1967 if (!ret)
1968 ret = -EPIPE;
1969 break;
1970 }
Jens Axboe70524492006-04-11 15:51:17 +02001971
1972 /*
Jens Axboeaadd06e2006-07-10 11:00:01 +02001973 * If we have iterated all input buffers or ran out of
1974 * output room, break.
Jens Axboe70524492006-04-11 15:51:17 +02001975 */
Jens Axboe35f3d142010-05-20 10:43:18 +02001976 if (i >= ipipe->nrbufs || opipe->nrbufs >= opipe->buffers)
Jens Axboe70524492006-04-11 15:51:17 +02001977 break;
Jens Axboeaadd06e2006-07-10 11:00:01 +02001978
Jens Axboe35f3d142010-05-20 10:43:18 +02001979 ibuf = ipipe->bufs + ((ipipe->curbuf + i) & (ipipe->buffers-1));
1980 nbuf = (opipe->curbuf + opipe->nrbufs) & (opipe->buffers - 1);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001981
Jens Axboe2a272502006-04-19 15:56:40 +02001982 /*
Jens Axboeaadd06e2006-07-10 11:00:01 +02001983 * Get a reference to this pipe buffer,
1984 * so we can copy the contents over.
Jens Axboe2a272502006-04-19 15:56:40 +02001985 */
Jens Axboeaadd06e2006-07-10 11:00:01 +02001986 ibuf->ops->get(ipipe, ibuf);
Jens Axboe70524492006-04-11 15:51:17 +02001987
Jens Axboeaadd06e2006-07-10 11:00:01 +02001988 obuf = opipe->bufs + nbuf;
1989 *obuf = *ibuf;
Jens Axboe70524492006-04-11 15:51:17 +02001990
Jens Axboeaadd06e2006-07-10 11:00:01 +02001991 /*
1992 * Don't inherit the gift flag, we need to
1993 * prevent multiple steals of this page.
1994 */
1995 obuf->flags &= ~PIPE_BUF_FLAG_GIFT;
1996
1997 if (obuf->len > len)
1998 obuf->len = len;
1999
2000 opipe->nrbufs++;
2001 ret += obuf->len;
2002 len -= obuf->len;
2003 i++;
2004 } while (len);
Jens Axboe70524492006-04-11 15:51:17 +02002005
Jens Axboe02cf01a2008-02-20 10:34:51 +01002006 /*
2007 * return EAGAIN if we have the potential of some data in the
2008 * future, otherwise just return 0
2009 */
2010 if (!ret && ipipe->waiting_writers && (flags & SPLICE_F_NONBLOCK))
2011 ret = -EAGAIN;
2012
Miklos Szeredi61e0d472009-04-14 19:48:41 +02002013 pipe_unlock(ipipe);
2014 pipe_unlock(opipe);
Jens Axboe70524492006-04-11 15:51:17 +02002015
Jens Axboeaadd06e2006-07-10 11:00:01 +02002016 /*
2017 * If we put data in the output pipe, wakeup any potential readers.
2018 */
2019 if (ret > 0) {
Jens Axboe70524492006-04-11 15:51:17 +02002020 smp_mb();
2021 if (waitqueue_active(&opipe->wait))
2022 wake_up_interruptible(&opipe->wait);
2023 kill_fasync(&opipe->fasync_readers, SIGIO, POLL_IN);
2024 }
2025
2026 return ret;
2027}
2028
2029/*
2030 * This is a tee(1) implementation that works on pipes. It doesn't copy
2031 * any data, it simply references the 'in' pages on the 'out' pipe.
2032 * The 'flags' used are the SPLICE_F_* variants, currently the only
2033 * applicable one is SPLICE_F_NONBLOCK.
2034 */
2035static long do_tee(struct file *in, struct file *out, size_t len,
2036 unsigned int flags)
2037{
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08002038 struct pipe_inode_info *ipipe = pipe_info(in->f_path.dentry->d_inode);
2039 struct pipe_inode_info *opipe = pipe_info(out->f_path.dentry->d_inode);
Jens Axboeaadd06e2006-07-10 11:00:01 +02002040 int ret = -EINVAL;
Jens Axboe70524492006-04-11 15:51:17 +02002041
2042 /*
Jens Axboeaadd06e2006-07-10 11:00:01 +02002043 * Duplicate the contents of ipipe to opipe without actually
2044 * copying the data.
Jens Axboe70524492006-04-11 15:51:17 +02002045 */
Jens Axboeaadd06e2006-07-10 11:00:01 +02002046 if (ipipe && opipe && ipipe != opipe) {
2047 /*
2048 * Keep going, unless we encounter an error. The ipipe/opipe
2049 * ordering doesn't really matter.
2050 */
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02002051 ret = ipipe_prep(ipipe, flags);
Jens Axboeaadd06e2006-07-10 11:00:01 +02002052 if (!ret) {
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02002053 ret = opipe_prep(opipe, flags);
Jens Axboe02cf01a2008-02-20 10:34:51 +01002054 if (!ret)
Jens Axboeaadd06e2006-07-10 11:00:01 +02002055 ret = link_pipe(ipipe, opipe, len, flags);
Jens Axboeaadd06e2006-07-10 11:00:01 +02002056 }
2057 }
Jens Axboe70524492006-04-11 15:51:17 +02002058
Jens Axboeaadd06e2006-07-10 11:00:01 +02002059 return ret;
Jens Axboe70524492006-04-11 15:51:17 +02002060}
2061
Heiko Carstens836f92a2009-01-14 14:14:33 +01002062SYSCALL_DEFINE4(tee, int, fdin, int, fdout, size_t, len, unsigned int, flags)
Jens Axboe70524492006-04-11 15:51:17 +02002063{
2064 struct file *in;
2065 int error, fput_in;
2066
2067 if (unlikely(!len))
2068 return 0;
2069
2070 error = -EBADF;
2071 in = fget_light(fdin, &fput_in);
2072 if (in) {
2073 if (in->f_mode & FMODE_READ) {
2074 int fput_out;
2075 struct file *out = fget_light(fdout, &fput_out);
2076
2077 if (out) {
2078 if (out->f_mode & FMODE_WRITE)
2079 error = do_tee(in, out, len, flags);
2080 fput_light(out, fput_out);
2081 }
2082 }
2083 fput_light(in, fput_in);
2084 }
2085
2086 return error;
2087}