blob: 8f1dfaecc8f06125429d3bc66b052d779f61f446 [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)) {
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;
Jens Axboef84d7512006-05-01 19:59:03 +0200685 int ret, more;
Jens Axboe5274f052006-03-30 15:15:30 +0200686
Jens Axboecac36bb2007-06-14 13:10:48 +0200687 ret = buf->ops->confirm(pipe, buf);
Jens Axboef84d7512006-05-01 19:59:03 +0200688 if (!ret) {
689 more = (sd->flags & SPLICE_F_MORE) || sd->len < sd->total_len;
Changli Gaocc56f7d2009-11-04 09:09:52 +0100690 if (file->f_op && file->f_op->sendpage)
691 ret = file->f_op->sendpage(file, buf->page, buf->offset,
692 sd->len, &pos, more);
693 else
694 ret = -EINVAL;
Jens Axboef84d7512006-05-01 19:59:03 +0200695 }
Jens Axboe5274f052006-03-30 15:15:30 +0200696
Jens Axboe016b6612006-04-25 15:42:00 +0200697 return ret;
Jens Axboe5274f052006-03-30 15:15:30 +0200698}
699
700/*
701 * This is a little more tricky than the file -> pipe splicing. There are
702 * basically three cases:
703 *
704 * - Destination page already exists in the address space and there
705 * are users of it. For that case we have no other option that
706 * copying the data. Tough luck.
707 * - Destination page already exists in the address space, but there
708 * are no users of it. Make sure it's uptodate, then drop it. Fall
709 * through to last case.
710 * - Destination page does not exist, we can add the pipe page to
711 * the page cache and avoid the copy.
712 *
Jens Axboe83f91352006-04-02 23:05:09 +0200713 * If asked to move pages to the output file (SPLICE_F_MOVE is set in
714 * sd->flags), we attempt to migrate pages from the pipe to the output
715 * file address space page cache. This is possible if no one else has
716 * the pipe page referenced outside of the pipe and page cache. If
717 * SPLICE_F_MOVE isn't set, or we cannot move the page, we simply create
718 * a new page in the output file page cache and fill/dirty that.
Jens Axboe5274f052006-03-30 15:15:30 +0200719 */
Miklos Szeredi328eaab2009-04-14 19:48:39 +0200720int pipe_to_file(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
721 struct splice_desc *sd)
Jens Axboe5274f052006-03-30 15:15:30 +0200722{
Jens Axboe6a14b902007-06-14 13:08:55 +0200723 struct file *file = sd->u.file;
Jens Axboe5274f052006-03-30 15:15:30 +0200724 struct address_space *mapping = file->f_mapping;
Jens Axboe016b6612006-04-25 15:42:00 +0200725 unsigned int offset, this_len;
Jens Axboe5274f052006-03-30 15:15:30 +0200726 struct page *page;
Nick Pigginafddba42007-10-16 01:25:01 -0700727 void *fsdata;
Jens Axboe3e7ee3e2006-04-02 23:11:04 +0200728 int ret;
Jens Axboe5274f052006-03-30 15:15:30 +0200729
730 /*
Jens Axboe49d0b212006-04-10 09:04:41 +0200731 * make sure the data in this buffer is uptodate
Jens Axboe5274f052006-03-30 15:15:30 +0200732 */
Jens Axboecac36bb2007-06-14 13:10:48 +0200733 ret = buf->ops->confirm(pipe, buf);
Jens Axboef84d7512006-05-01 19:59:03 +0200734 if (unlikely(ret))
735 return ret;
Jens Axboe5274f052006-03-30 15:15:30 +0200736
Jens Axboe5274f052006-03-30 15:15:30 +0200737 offset = sd->pos & ~PAGE_CACHE_MASK;
738
Jens Axboe016b6612006-04-25 15:42:00 +0200739 this_len = sd->len;
740 if (this_len + offset > PAGE_CACHE_SIZE)
741 this_len = PAGE_CACHE_SIZE - offset;
742
Nick Pigginafddba42007-10-16 01:25:01 -0700743 ret = pagecache_write_begin(file, mapping, sd->pos, this_len,
744 AOP_FLAG_UNINTERRUPTIBLE, &page, &fsdata);
745 if (unlikely(ret))
746 goto out;
Jens Axboe5274f052006-03-30 15:15:30 +0200747
Jens Axboe0568b402006-05-01 19:50:48 +0200748 if (buf->page != page) {
Jens Axboef84d7512006-05-01 19:59:03 +0200749 /*
750 * Careful, ->map() uses KM_USER0!
751 */
Jens Axboe76ad4d12006-05-03 10:41:33 +0200752 char *src = buf->ops->map(pipe, buf, 1);
Jens Axboef84d7512006-05-01 19:59:03 +0200753 char *dst = kmap_atomic(page, KM_USER1);
Jens Axboe5abc97a2006-03-30 15:16:46 +0200754
Jens Axboe016b6612006-04-25 15:42:00 +0200755 memcpy(dst + offset, src + buf->offset, this_len);
Jens Axboe5abc97a2006-03-30 15:16:46 +0200756 flush_dcache_page(page);
Jens Axboef84d7512006-05-01 19:59:03 +0200757 kunmap_atomic(dst, KM_USER1);
Jens Axboe76ad4d12006-05-03 10:41:33 +0200758 buf->ops->unmap(pipe, buf, src);
Jens Axboe5abc97a2006-03-30 15:16:46 +0200759 }
Nick Pigginafddba42007-10-16 01:25:01 -0700760 ret = pagecache_write_end(file, mapping, sd->pos, this_len, this_len,
761 page, fsdata);
Jens Axboe5274f052006-03-30 15:15:30 +0200762out:
Jens Axboe5274f052006-03-30 15:15:30 +0200763 return ret;
764}
Miklos Szeredi328eaab2009-04-14 19:48:39 +0200765EXPORT_SYMBOL(pipe_to_file);
Jens Axboe5274f052006-03-30 15:15:30 +0200766
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200767static void wakeup_pipe_writers(struct pipe_inode_info *pipe)
768{
769 smp_mb();
770 if (waitqueue_active(&pipe->wait))
771 wake_up_interruptible(&pipe->wait);
772 kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
773}
774
775/**
776 * splice_from_pipe_feed - feed available data from a pipe to a file
777 * @pipe: pipe to splice from
778 * @sd: information to @actor
779 * @actor: handler that splices the data
780 *
781 * Description:
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200782 * This function loops over the pipe and calls @actor to do the
783 * actual moving of a single struct pipe_buffer to the desired
784 * destination. It returns when there's no more buffers left in
785 * the pipe or if the requested number of bytes (@sd->total_len)
786 * have been copied. It returns a positive number (one) if the
787 * pipe needs to be filled with more data, zero if the required
788 * number of bytes have been copied and -errno on error.
789 *
790 * This, together with splice_from_pipe_{begin,end,next}, may be
791 * used to implement the functionality of __splice_from_pipe() when
792 * locking is required around copying the pipe buffers to the
793 * destination.
794 */
795int splice_from_pipe_feed(struct pipe_inode_info *pipe, struct splice_desc *sd,
796 splice_actor *actor)
797{
798 int ret;
799
800 while (pipe->nrbufs) {
801 struct pipe_buffer *buf = pipe->bufs + pipe->curbuf;
802 const struct pipe_buf_operations *ops = buf->ops;
803
804 sd->len = buf->len;
805 if (sd->len > sd->total_len)
806 sd->len = sd->total_len;
807
808 ret = actor(pipe, buf, sd);
809 if (ret <= 0) {
810 if (ret == -ENODATA)
811 ret = 0;
812 return ret;
813 }
814 buf->offset += ret;
815 buf->len -= ret;
816
817 sd->num_spliced += ret;
818 sd->len -= ret;
819 sd->pos += ret;
820 sd->total_len -= ret;
821
822 if (!buf->len) {
823 buf->ops = NULL;
824 ops->release(pipe, buf);
Jens Axboe35f3d142010-05-20 10:43:18 +0200825 pipe->curbuf = (pipe->curbuf + 1) & (pipe->buffers - 1);
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200826 pipe->nrbufs--;
827 if (pipe->inode)
828 sd->need_wakeup = true;
829 }
830
831 if (!sd->total_len)
832 return 0;
833 }
834
835 return 1;
836}
837EXPORT_SYMBOL(splice_from_pipe_feed);
838
839/**
840 * splice_from_pipe_next - wait for some data to splice from
841 * @pipe: pipe to splice from
842 * @sd: information about the splice operation
843 *
844 * Description:
845 * This function will wait for some data and return a positive
846 * value (one) if pipe buffers are available. It will return zero
847 * or -errno if no more data needs to be spliced.
848 */
849int splice_from_pipe_next(struct pipe_inode_info *pipe, struct splice_desc *sd)
850{
851 while (!pipe->nrbufs) {
852 if (!pipe->writers)
853 return 0;
854
855 if (!pipe->waiting_writers && sd->num_spliced)
856 return 0;
857
858 if (sd->flags & SPLICE_F_NONBLOCK)
859 return -EAGAIN;
860
861 if (signal_pending(current))
862 return -ERESTARTSYS;
863
864 if (sd->need_wakeup) {
865 wakeup_pipe_writers(pipe);
866 sd->need_wakeup = false;
867 }
868
869 pipe_wait(pipe);
870 }
871
872 return 1;
873}
874EXPORT_SYMBOL(splice_from_pipe_next);
875
876/**
877 * splice_from_pipe_begin - start splicing from pipe
Randy Dunlapb80901b2009-04-16 19:09:55 -0700878 * @sd: information about the splice operation
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200879 *
880 * Description:
881 * This function should be called before a loop containing
882 * splice_from_pipe_next() and splice_from_pipe_feed() to
883 * initialize the necessary fields of @sd.
884 */
885void splice_from_pipe_begin(struct splice_desc *sd)
886{
887 sd->num_spliced = 0;
888 sd->need_wakeup = false;
889}
890EXPORT_SYMBOL(splice_from_pipe_begin);
891
892/**
893 * splice_from_pipe_end - finish splicing from pipe
894 * @pipe: pipe to splice from
895 * @sd: information about the splice operation
896 *
897 * Description:
898 * This function will wake up pipe writers if necessary. It should
899 * be called after a loop containing splice_from_pipe_next() and
900 * splice_from_pipe_feed().
901 */
902void splice_from_pipe_end(struct pipe_inode_info *pipe, struct splice_desc *sd)
903{
904 if (sd->need_wakeup)
905 wakeup_pipe_writers(pipe);
906}
907EXPORT_SYMBOL(splice_from_pipe_end);
908
Jens Axboe932cc6d2007-06-21 13:10:21 +0200909/**
910 * __splice_from_pipe - splice data from a pipe to given actor
911 * @pipe: pipe to splice from
912 * @sd: information to @actor
913 * @actor: handler that splices the data
914 *
915 * Description:
916 * This function does little more than loop over the pipe and call
917 * @actor to do the actual moving of a single struct pipe_buffer to
918 * the desired destination. See pipe_to_file, pipe_to_sendpage, or
919 * pipe_to_user.
920 *
Jens Axboe83f91352006-04-02 23:05:09 +0200921 */
Jens Axboec66ab6f2007-06-12 21:17:17 +0200922ssize_t __splice_from_pipe(struct pipe_inode_info *pipe, struct splice_desc *sd,
923 splice_actor *actor)
Jens Axboe5274f052006-03-30 15:15:30 +0200924{
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200925 int ret;
Jens Axboe5274f052006-03-30 15:15:30 +0200926
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200927 splice_from_pipe_begin(sd);
928 do {
929 ret = splice_from_pipe_next(pipe, sd);
930 if (ret > 0)
931 ret = splice_from_pipe_feed(pipe, sd, actor);
932 } while (ret > 0);
933 splice_from_pipe_end(pipe, sd);
Jens Axboe5274f052006-03-30 15:15:30 +0200934
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200935 return sd->num_spliced ? sd->num_spliced : ret;
Jens Axboe5274f052006-03-30 15:15:30 +0200936}
Mark Fasheh40bee442007-03-21 13:11:02 +0100937EXPORT_SYMBOL(__splice_from_pipe);
Jens Axboe5274f052006-03-30 15:15:30 +0200938
Jens Axboe932cc6d2007-06-21 13:10:21 +0200939/**
940 * splice_from_pipe - splice data from a pipe to a file
941 * @pipe: pipe to splice from
942 * @out: file to splice to
943 * @ppos: position in @out
944 * @len: how many bytes to splice
945 * @flags: splice modifier flags
946 * @actor: handler that splices the data
947 *
948 * Description:
Miklos Szeredi29339702009-04-14 19:48:37 +0200949 * See __splice_from_pipe. This function locks the pipe inode,
Jens Axboe932cc6d2007-06-21 13:10:21 +0200950 * otherwise it's identical to __splice_from_pipe().
951 *
952 */
Mark Fasheh6da61802006-10-17 18:43:07 +0200953ssize_t splice_from_pipe(struct pipe_inode_info *pipe, struct file *out,
954 loff_t *ppos, size_t len, unsigned int flags,
955 splice_actor *actor)
956{
957 ssize_t ret;
Jens Axboec66ab6f2007-06-12 21:17:17 +0200958 struct splice_desc sd = {
959 .total_len = len,
960 .flags = flags,
961 .pos = *ppos,
Jens Axboe6a14b902007-06-14 13:08:55 +0200962 .u.file = out,
Jens Axboec66ab6f2007-06-12 21:17:17 +0200963 };
Mark Fasheh6da61802006-10-17 18:43:07 +0200964
Miklos Szeredi61e0d472009-04-14 19:48:41 +0200965 pipe_lock(pipe);
Jens Axboec66ab6f2007-06-12 21:17:17 +0200966 ret = __splice_from_pipe(pipe, &sd, actor);
Miklos Szeredi61e0d472009-04-14 19:48:41 +0200967 pipe_unlock(pipe);
Mark Fasheh6da61802006-10-17 18:43:07 +0200968
969 return ret;
970}
971
972/**
Jens Axboe83f91352006-04-02 23:05:09 +0200973 * generic_file_splice_write - splice data from a pipe to a file
Ingo Molnar3a326a22006-04-10 15:18:35 +0200974 * @pipe: pipe info
Jens Axboe83f91352006-04-02 23:05:09 +0200975 * @out: file to write to
Jens Axboe932cc6d2007-06-21 13:10:21 +0200976 * @ppos: position in @out
Jens Axboe83f91352006-04-02 23:05:09 +0200977 * @len: number of bytes to splice
978 * @flags: splice modifier flags
979 *
Jens Axboe932cc6d2007-06-21 13:10:21 +0200980 * Description:
981 * Will either move or copy pages (determined by @flags options) from
982 * the given pipe inode to the given file.
Jens Axboe83f91352006-04-02 23:05:09 +0200983 *
984 */
Ingo Molnar3a326a22006-04-10 15:18:35 +0200985ssize_t
986generic_file_splice_write(struct pipe_inode_info *pipe, struct file *out,
Jens Axboecbb7e572006-04-11 14:57:50 +0200987 loff_t *ppos, size_t len, unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +0200988{
Jens Axboe4f6f0bd2006-04-02 23:04:46 +0200989 struct address_space *mapping = out->f_mapping;
Jens Axboe8c34e2d2006-10-17 19:43:22 +0200990 struct inode *inode = mapping->host;
Miklos Szeredi7f3d4ee2008-05-07 09:22:39 +0200991 struct splice_desc sd = {
992 .total_len = len,
993 .flags = flags,
994 .pos = *ppos,
995 .u.file = out,
996 };
Ingo Molnar3a326a22006-04-10 15:18:35 +0200997 ssize_t ret;
Jens Axboe8c34e2d2006-10-17 19:43:22 +0200998
Miklos Szeredi61e0d472009-04-14 19:48:41 +0200999 pipe_lock(pipe);
Miklos Szeredieb443e52009-04-14 19:48:38 +02001000
1001 splice_from_pipe_begin(&sd);
1002 do {
1003 ret = splice_from_pipe_next(pipe, &sd);
1004 if (ret <= 0)
1005 break;
1006
1007 mutex_lock_nested(&inode->i_mutex, I_MUTEX_CHILD);
1008 ret = file_remove_suid(out);
Miklos Szeredi723590e2009-08-15 08:43:22 +02001009 if (!ret) {
1010 file_update_time(out);
Miklos Szeredieb443e52009-04-14 19:48:38 +02001011 ret = splice_from_pipe_feed(pipe, &sd, pipe_to_file);
Miklos Szeredi723590e2009-08-15 08:43:22 +02001012 }
Miklos Szeredieb443e52009-04-14 19:48:38 +02001013 mutex_unlock(&inode->i_mutex);
1014 } while (ret > 0);
1015 splice_from_pipe_end(pipe, &sd);
1016
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001017 pipe_unlock(pipe);
Miklos Szeredieb443e52009-04-14 19:48:38 +02001018
1019 if (sd.num_spliced)
1020 ret = sd.num_spliced;
1021
Jens Axboea4514eb2006-04-19 15:57:05 +02001022 if (ret > 0) {
Jens Axboe17ee4f42007-06-15 13:10:37 +02001023 unsigned long nr_pages;
Jan Kara148f9482009-08-17 19:52:36 +02001024 int err;
Jens Axboe17ee4f42007-06-15 13:10:37 +02001025
Jens Axboe17ee4f42007-06-15 13:10:37 +02001026 nr_pages = (ret + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
Jens Axboe4f6f0bd2006-04-02 23:04:46 +02001027
Jan Kara148f9482009-08-17 19:52:36 +02001028 err = generic_write_sync(out, *ppos, ret);
1029 if (err)
1030 ret = err;
1031 else
1032 *ppos += ret;
Jens Axboe17ee4f42007-06-15 13:10:37 +02001033 balance_dirty_pages_ratelimited_nr(mapping, nr_pages);
Jens Axboe4f6f0bd2006-04-02 23:04:46 +02001034 }
1035
1036 return ret;
Jens Axboe5274f052006-03-30 15:15:30 +02001037}
1038
Jens Axboe059a8f32006-04-02 23:06:05 +02001039EXPORT_SYMBOL(generic_file_splice_write);
1040
Miklos Szeredib2858d72009-05-19 11:37:46 +02001041static int write_pipe_buf(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
1042 struct splice_desc *sd)
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +02001043{
Miklos Szeredib2858d72009-05-19 11:37:46 +02001044 int ret;
1045 void *data;
1046
1047 ret = buf->ops->confirm(pipe, buf);
1048 if (ret)
1049 return ret;
1050
1051 data = buf->ops->map(pipe, buf, 0);
1052 ret = kernel_write(sd->u.file, data + buf->offset, sd->len, sd->pos);
1053 buf->ops->unmap(pipe, buf, data);
1054
1055 return ret;
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +02001056}
1057
1058static ssize_t default_file_splice_write(struct pipe_inode_info *pipe,
1059 struct file *out, loff_t *ppos,
1060 size_t len, unsigned int flags)
1061{
Miklos Szeredib2858d72009-05-19 11:37:46 +02001062 ssize_t ret;
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +02001063
Miklos Szeredib2858d72009-05-19 11:37:46 +02001064 ret = splice_from_pipe(pipe, out, ppos, len, flags, write_pipe_buf);
1065 if (ret > 0)
1066 *ppos += ret;
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +02001067
Miklos Szeredib2858d72009-05-19 11:37:46 +02001068 return ret;
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +02001069}
1070
Jens Axboe83f91352006-04-02 23:05:09 +02001071/**
1072 * generic_splice_sendpage - splice data from a pipe to a socket
Jens Axboe932cc6d2007-06-21 13:10:21 +02001073 * @pipe: pipe to splice from
Jens Axboe83f91352006-04-02 23:05:09 +02001074 * @out: socket to write to
Jens Axboe932cc6d2007-06-21 13:10:21 +02001075 * @ppos: position in @out
Jens Axboe83f91352006-04-02 23:05:09 +02001076 * @len: number of bytes to splice
1077 * @flags: splice modifier flags
1078 *
Jens Axboe932cc6d2007-06-21 13:10:21 +02001079 * Description:
1080 * Will send @len bytes from the pipe to a network socket. No data copying
1081 * is involved.
Jens Axboe83f91352006-04-02 23:05:09 +02001082 *
1083 */
Ingo Molnar3a326a22006-04-10 15:18:35 +02001084ssize_t generic_splice_sendpage(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{
Jens Axboe00522fb2006-04-26 14:39:29 +02001087 return splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_sendpage);
Jens Axboe5274f052006-03-30 15:15:30 +02001088}
1089
Jens Axboe059a8f32006-04-02 23:06:05 +02001090EXPORT_SYMBOL(generic_splice_sendpage);
Jeff Garzika0f06782006-03-30 23:06:13 -05001091
Jens Axboe83f91352006-04-02 23:05:09 +02001092/*
1093 * Attempt to initiate a splice from pipe to file.
1094 */
Ingo Molnar3a326a22006-04-10 15:18:35 +02001095static long do_splice_from(struct pipe_inode_info *pipe, struct file *out,
Jens Axboecbb7e572006-04-11 14:57:50 +02001096 loff_t *ppos, size_t len, unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +02001097{
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +02001098 ssize_t (*splice_write)(struct pipe_inode_info *, struct file *,
1099 loff_t *, size_t, unsigned int);
Jens Axboe5274f052006-03-30 15:15:30 +02001100 int ret;
1101
Jens Axboe49570e92006-04-11 13:56:09 +02001102 if (unlikely(!(out->f_mode & FMODE_WRITE)))
Jens Axboe5274f052006-03-30 15:15:30 +02001103 return -EBADF;
1104
Linus Torvaldsefc968d2008-10-09 14:04:54 -07001105 if (unlikely(out->f_flags & O_APPEND))
1106 return -EINVAL;
1107
Jens Axboecbb7e572006-04-11 14:57:50 +02001108 ret = rw_verify_area(WRITE, out, ppos, len);
Jens Axboe5274f052006-03-30 15:15:30 +02001109 if (unlikely(ret < 0))
1110 return ret;
1111
Changli Gaocc56f7d2009-11-04 09:09:52 +01001112 if (out->f_op && out->f_op->splice_write)
1113 splice_write = out->f_op->splice_write;
1114 else
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +02001115 splice_write = default_file_splice_write;
1116
1117 return splice_write(pipe, out, ppos, len, flags);
Jens Axboe5274f052006-03-30 15:15:30 +02001118}
1119
Jens Axboe83f91352006-04-02 23:05:09 +02001120/*
1121 * Attempt to initiate a splice from a file to a pipe.
1122 */
Jens Axboecbb7e572006-04-11 14:57:50 +02001123static long do_splice_to(struct file *in, loff_t *ppos,
1124 struct pipe_inode_info *pipe, size_t len,
1125 unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +02001126{
Miklos Szeredi68181732009-05-07 15:37:36 +02001127 ssize_t (*splice_read)(struct file *, loff_t *,
1128 struct pipe_inode_info *, size_t, unsigned int);
Jens Axboe5274f052006-03-30 15:15:30 +02001129 int ret;
1130
Jens Axboe49570e92006-04-11 13:56:09 +02001131 if (unlikely(!(in->f_mode & FMODE_READ)))
Jens Axboe5274f052006-03-30 15:15:30 +02001132 return -EBADF;
1133
Jens Axboecbb7e572006-04-11 14:57:50 +02001134 ret = rw_verify_area(READ, in, ppos, len);
Jens Axboe5274f052006-03-30 15:15:30 +02001135 if (unlikely(ret < 0))
1136 return ret;
1137
Changli Gaocc56f7d2009-11-04 09:09:52 +01001138 if (in->f_op && in->f_op->splice_read)
1139 splice_read = in->f_op->splice_read;
1140 else
Miklos Szeredi68181732009-05-07 15:37:36 +02001141 splice_read = default_file_splice_read;
1142
1143 return splice_read(in, ppos, pipe, len, flags);
Jens Axboe5274f052006-03-30 15:15:30 +02001144}
1145
Jens Axboe932cc6d2007-06-21 13:10:21 +02001146/**
1147 * splice_direct_to_actor - splices data directly between two non-pipes
1148 * @in: file to splice from
1149 * @sd: actor information on where to splice to
1150 * @actor: handles the data splicing
1151 *
1152 * Description:
1153 * This is a special case helper to splice directly between two
1154 * points, without requiring an explicit pipe. Internally an allocated
Randy Dunlap79685b82007-07-27 08:08:51 +02001155 * pipe is cached in the process, and reused during the lifetime of
Jens Axboe932cc6d2007-06-21 13:10:21 +02001156 * that process.
1157 *
Jens Axboec66ab6f2007-06-12 21:17:17 +02001158 */
1159ssize_t splice_direct_to_actor(struct file *in, struct splice_desc *sd,
1160 splice_direct_actor *actor)
Jens Axboeb92ce552006-04-11 13:52:07 +02001161{
1162 struct pipe_inode_info *pipe;
1163 long ret, bytes;
1164 umode_t i_mode;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001165 size_t len;
1166 int i, flags;
Jens Axboeb92ce552006-04-11 13:52:07 +02001167
1168 /*
1169 * We require the input being a regular file, as we don't want to
1170 * randomly drop data for eg socket -> socket splicing. Use the
1171 * piped splicing for that!
1172 */
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001173 i_mode = in->f_path.dentry->d_inode->i_mode;
Jens Axboeb92ce552006-04-11 13:52:07 +02001174 if (unlikely(!S_ISREG(i_mode) && !S_ISBLK(i_mode)))
1175 return -EINVAL;
1176
1177 /*
1178 * neither in nor out is a pipe, setup an internal pipe attached to
1179 * 'out' and transfer the wanted data from 'in' to 'out' through that
1180 */
1181 pipe = current->splice_pipe;
Jens Axboe49570e92006-04-11 13:56:09 +02001182 if (unlikely(!pipe)) {
Jens Axboeb92ce552006-04-11 13:52:07 +02001183 pipe = alloc_pipe_info(NULL);
1184 if (!pipe)
1185 return -ENOMEM;
1186
1187 /*
1188 * We don't have an immediate reader, but we'll read the stuff
Jens Axboe00522fb2006-04-26 14:39:29 +02001189 * out of the pipe right after the splice_to_pipe(). So set
Jens Axboeb92ce552006-04-11 13:52:07 +02001190 * PIPE_READERS appropriately.
1191 */
1192 pipe->readers = 1;
1193
1194 current->splice_pipe = pipe;
1195 }
1196
1197 /*
Ingo Molnar73d62d82006-04-11 13:57:21 +02001198 * Do the splice.
Jens Axboeb92ce552006-04-11 13:52:07 +02001199 */
1200 ret = 0;
1201 bytes = 0;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001202 len = sd->total_len;
1203 flags = sd->flags;
1204
1205 /*
1206 * Don't block on output, we have to drain the direct pipe.
1207 */
1208 sd->flags &= ~SPLICE_F_NONBLOCK;
Jens Axboeb92ce552006-04-11 13:52:07 +02001209
1210 while (len) {
Jens Axboe51a92c02007-07-13 14:11:43 +02001211 size_t read_len;
Tom Zanussia82c53a2008-05-09 13:28:36 +02001212 loff_t pos = sd->pos, prev_pos = pos;
Jens Axboeb92ce552006-04-11 13:52:07 +02001213
Jens Axboebcd4f3a2007-07-16 14:41:49 +02001214 ret = do_splice_to(in, &pos, pipe, len, flags);
Jens Axboe51a92c02007-07-13 14:11:43 +02001215 if (unlikely(ret <= 0))
Jens Axboeb92ce552006-04-11 13:52:07 +02001216 goto out_release;
1217
1218 read_len = ret;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001219 sd->total_len = read_len;
Jens Axboeb92ce552006-04-11 13:52:07 +02001220
1221 /*
1222 * NOTE: nonblocking mode only applies to the input. We
1223 * must not do the output in nonblocking mode as then we
1224 * could get stuck data in the internal pipe:
1225 */
Jens Axboec66ab6f2007-06-12 21:17:17 +02001226 ret = actor(pipe, sd);
Tom Zanussia82c53a2008-05-09 13:28:36 +02001227 if (unlikely(ret <= 0)) {
1228 sd->pos = prev_pos;
Jens Axboeb92ce552006-04-11 13:52:07 +02001229 goto out_release;
Tom Zanussia82c53a2008-05-09 13:28:36 +02001230 }
Jens Axboeb92ce552006-04-11 13:52:07 +02001231
1232 bytes += ret;
1233 len -= ret;
Jens Axboebcd4f3a2007-07-16 14:41:49 +02001234 sd->pos = pos;
Jens Axboeb92ce552006-04-11 13:52:07 +02001235
Tom Zanussia82c53a2008-05-09 13:28:36 +02001236 if (ret < read_len) {
1237 sd->pos = prev_pos + ret;
Jens Axboe51a92c02007-07-13 14:11:43 +02001238 goto out_release;
Tom Zanussia82c53a2008-05-09 13:28:36 +02001239 }
Jens Axboeb92ce552006-04-11 13:52:07 +02001240 }
1241
Jens Axboe9e971982008-01-29 21:05:57 +01001242done:
Jens Axboeb92ce552006-04-11 13:52:07 +02001243 pipe->nrbufs = pipe->curbuf = 0;
Jens Axboe80848702008-01-30 12:24:48 +01001244 file_accessed(in);
Jens Axboeb92ce552006-04-11 13:52:07 +02001245 return bytes;
1246
1247out_release:
1248 /*
1249 * If we did an incomplete transfer we must release
1250 * the pipe buffers in question:
1251 */
Jens Axboe35f3d142010-05-20 10:43:18 +02001252 for (i = 0; i < pipe->buffers; i++) {
Jens Axboeb92ce552006-04-11 13:52:07 +02001253 struct pipe_buffer *buf = pipe->bufs + i;
1254
1255 if (buf->ops) {
1256 buf->ops->release(pipe, buf);
1257 buf->ops = NULL;
1258 }
1259 }
Jens Axboeb92ce552006-04-11 13:52:07 +02001260
Jens Axboe9e971982008-01-29 21:05:57 +01001261 if (!bytes)
1262 bytes = ret;
Jens Axboeb92ce552006-04-11 13:52:07 +02001263
Jens Axboe9e971982008-01-29 21:05:57 +01001264 goto done;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001265}
1266EXPORT_SYMBOL(splice_direct_to_actor);
1267
1268static int direct_splice_actor(struct pipe_inode_info *pipe,
1269 struct splice_desc *sd)
1270{
Jens Axboe6a14b902007-06-14 13:08:55 +02001271 struct file *file = sd->u.file;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001272
Changli Gao2cb4b052010-06-29 13:09:18 +02001273 return do_splice_from(pipe, file, &file->f_pos, sd->total_len,
1274 sd->flags);
Jens Axboec66ab6f2007-06-12 21:17:17 +02001275}
1276
Jens Axboe932cc6d2007-06-21 13:10:21 +02001277/**
1278 * do_splice_direct - splices data directly between two files
1279 * @in: file to splice from
1280 * @ppos: input file offset
1281 * @out: file to splice to
1282 * @len: number of bytes to splice
1283 * @flags: splice modifier flags
1284 *
1285 * Description:
1286 * For use by do_sendfile(). splice can easily emulate sendfile, but
1287 * doing it in the application would incur an extra system call
1288 * (splice in + splice out, as compared to just sendfile()). So this helper
1289 * can splice directly through a process-private pipe.
1290 *
1291 */
Jens Axboec66ab6f2007-06-12 21:17:17 +02001292long do_splice_direct(struct file *in, loff_t *ppos, struct file *out,
1293 size_t len, unsigned int flags)
1294{
1295 struct splice_desc sd = {
1296 .len = len,
1297 .total_len = len,
1298 .flags = flags,
1299 .pos = *ppos,
Jens Axboe6a14b902007-06-14 13:08:55 +02001300 .u.file = out,
Jens Axboec66ab6f2007-06-12 21:17:17 +02001301 };
Jens Axboe51a92c02007-07-13 14:11:43 +02001302 long ret;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001303
1304 ret = splice_direct_to_actor(in, &sd, direct_splice_actor);
Jens Axboe51a92c02007-07-13 14:11:43 +02001305 if (ret > 0)
Tom Zanussia82c53a2008-05-09 13:28:36 +02001306 *ppos = sd.pos;
Jens Axboe51a92c02007-07-13 14:11:43 +02001307
Jens Axboec66ab6f2007-06-12 21:17:17 +02001308 return ret;
Jens Axboeb92ce552006-04-11 13:52:07 +02001309}
1310
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001311static int splice_pipe_to_pipe(struct pipe_inode_info *ipipe,
1312 struct pipe_inode_info *opipe,
1313 size_t len, unsigned int flags);
Jens Axboe83f91352006-04-02 23:05:09 +02001314/*
Jens Axboeddac0d32006-11-04 12:49:32 +01001315 * After the inode slimming patch, i_pipe/i_bdev/i_cdev share the same
1316 * location, so checking ->i_pipe is not enough to verify that this is a
1317 * pipe.
1318 */
1319static inline struct pipe_inode_info *pipe_info(struct inode *inode)
1320{
1321 if (S_ISFIFO(inode->i_mode))
1322 return inode->i_pipe;
1323
1324 return NULL;
1325}
1326
1327/*
Jens Axboe83f91352006-04-02 23:05:09 +02001328 * Determine where to splice to/from.
1329 */
Ingo Molnar529565d2006-04-10 15:18:58 +02001330static long do_splice(struct file *in, loff_t __user *off_in,
1331 struct file *out, loff_t __user *off_out,
1332 size_t len, unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +02001333{
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001334 struct pipe_inode_info *ipipe;
1335 struct pipe_inode_info *opipe;
Jens Axboecbb7e572006-04-11 14:57:50 +02001336 loff_t offset, *off;
Jens Axboea4514eb2006-04-19 15:57:05 +02001337 long ret;
Jens Axboe5274f052006-03-30 15:15:30 +02001338
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001339 ipipe = pipe_info(in->f_path.dentry->d_inode);
1340 opipe = pipe_info(out->f_path.dentry->d_inode);
1341
1342 if (ipipe && opipe) {
1343 if (off_in || off_out)
1344 return -ESPIPE;
1345
1346 if (!(in->f_mode & FMODE_READ))
1347 return -EBADF;
1348
1349 if (!(out->f_mode & FMODE_WRITE))
1350 return -EBADF;
1351
1352 /* Splicing to self would be fun, but... */
1353 if (ipipe == opipe)
1354 return -EINVAL;
1355
1356 return splice_pipe_to_pipe(ipipe, opipe, len, flags);
1357 }
1358
1359 if (ipipe) {
Ingo Molnar529565d2006-04-10 15:18:58 +02001360 if (off_in)
1361 return -ESPIPE;
Jens Axboeb92ce552006-04-11 13:52:07 +02001362 if (off_out) {
Changli Gao19c9a492010-06-29 13:10:36 +02001363 if (!(out->f_mode & FMODE_PWRITE))
Jens Axboeb92ce552006-04-11 13:52:07 +02001364 return -EINVAL;
Jens Axboecbb7e572006-04-11 14:57:50 +02001365 if (copy_from_user(&offset, off_out, sizeof(loff_t)))
Jens Axboeb92ce552006-04-11 13:52:07 +02001366 return -EFAULT;
Jens Axboecbb7e572006-04-11 14:57:50 +02001367 off = &offset;
1368 } else
1369 off = &out->f_pos;
Ingo Molnar529565d2006-04-10 15:18:58 +02001370
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001371 ret = do_splice_from(ipipe, out, off, len, flags);
Jens Axboea4514eb2006-04-19 15:57:05 +02001372
1373 if (off_out && copy_to_user(off_out, off, sizeof(loff_t)))
1374 ret = -EFAULT;
1375
1376 return ret;
Ingo Molnar529565d2006-04-10 15:18:58 +02001377 }
Jens Axboe5274f052006-03-30 15:15:30 +02001378
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001379 if (opipe) {
Ingo Molnar529565d2006-04-10 15:18:58 +02001380 if (off_out)
1381 return -ESPIPE;
Jens Axboeb92ce552006-04-11 13:52:07 +02001382 if (off_in) {
Changli Gao19c9a492010-06-29 13:10:36 +02001383 if (!(in->f_mode & FMODE_PREAD))
Jens Axboeb92ce552006-04-11 13:52:07 +02001384 return -EINVAL;
Jens Axboecbb7e572006-04-11 14:57:50 +02001385 if (copy_from_user(&offset, off_in, sizeof(loff_t)))
Jens Axboeb92ce552006-04-11 13:52:07 +02001386 return -EFAULT;
Jens Axboecbb7e572006-04-11 14:57:50 +02001387 off = &offset;
1388 } else
1389 off = &in->f_pos;
Ingo Molnar529565d2006-04-10 15:18:58 +02001390
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001391 ret = do_splice_to(in, off, opipe, len, flags);
Jens Axboea4514eb2006-04-19 15:57:05 +02001392
1393 if (off_in && copy_to_user(off_in, off, sizeof(loff_t)))
1394 ret = -EFAULT;
1395
1396 return ret;
Ingo Molnar529565d2006-04-10 15:18:58 +02001397 }
Jens Axboe5274f052006-03-30 15:15:30 +02001398
1399 return -EINVAL;
1400}
1401
Jens Axboe912d35f2006-04-26 10:59:21 +02001402/*
1403 * Map an iov into an array of pages and offset/length tupples. With the
1404 * partial_page structure, we can map several non-contiguous ranges into
1405 * our ones pages[] map instead of splitting that operation into pieces.
1406 * Could easily be exported as a generic helper for other users, in which
1407 * case one would probably want to add a 'max_nr_pages' parameter as well.
1408 */
1409static int get_iovec_page_array(const struct iovec __user *iov,
1410 unsigned int nr_vecs, struct page **pages,
Jens Axboe35f3d142010-05-20 10:43:18 +02001411 struct partial_page *partial, int aligned,
1412 unsigned int pipe_buffers)
Jens Axboe912d35f2006-04-26 10:59:21 +02001413{
1414 int buffers = 0, error = 0;
1415
Jens Axboe912d35f2006-04-26 10:59:21 +02001416 while (nr_vecs) {
1417 unsigned long off, npages;
Linus Torvalds75723952007-10-01 13:17:28 -07001418 struct iovec entry;
Jens Axboe912d35f2006-04-26 10:59:21 +02001419 void __user *base;
1420 size_t len;
1421 int i;
1422
Linus Torvalds75723952007-10-01 13:17:28 -07001423 error = -EFAULT;
Nick Pigginbc40d732008-07-25 19:45:26 -07001424 if (copy_from_user(&entry, iov, sizeof(entry)))
Jens Axboe912d35f2006-04-26 10:59:21 +02001425 break;
Linus Torvalds75723952007-10-01 13:17:28 -07001426
1427 base = entry.iov_base;
1428 len = entry.iov_len;
Jens Axboe912d35f2006-04-26 10:59:21 +02001429
1430 /*
1431 * Sanity check this iovec. 0 read succeeds.
1432 */
Linus Torvalds75723952007-10-01 13:17:28 -07001433 error = 0;
Jens Axboe912d35f2006-04-26 10:59:21 +02001434 if (unlikely(!len))
1435 break;
1436 error = -EFAULT;
Bastian Blank712a30e2008-02-10 16:47:57 +02001437 if (!access_ok(VERIFY_READ, base, len))
Jens Axboe912d35f2006-04-26 10:59:21 +02001438 break;
1439
1440 /*
1441 * Get this base offset and number of pages, then map
1442 * in the user pages.
1443 */
1444 off = (unsigned long) base & ~PAGE_MASK;
Jens Axboe7afa6fd2006-05-01 20:02:33 +02001445
1446 /*
1447 * If asked for alignment, the offset must be zero and the
1448 * length a multiple of the PAGE_SIZE.
1449 */
1450 error = -EINVAL;
1451 if (aligned && (off || len & ~PAGE_MASK))
1452 break;
1453
Jens Axboe912d35f2006-04-26 10:59:21 +02001454 npages = (off + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
Jens Axboe35f3d142010-05-20 10:43:18 +02001455 if (npages > pipe_buffers - buffers)
1456 npages = pipe_buffers - buffers;
Jens Axboe912d35f2006-04-26 10:59:21 +02001457
Nick Pigginbc40d732008-07-25 19:45:26 -07001458 error = get_user_pages_fast((unsigned long)base, npages,
1459 0, &pages[buffers]);
Jens Axboe912d35f2006-04-26 10:59:21 +02001460
1461 if (unlikely(error <= 0))
1462 break;
1463
1464 /*
1465 * Fill this contiguous range into the partial page map.
1466 */
1467 for (i = 0; i < error; i++) {
Jens Axboe75914892006-05-02 12:57:18 +02001468 const int plen = min_t(size_t, len, PAGE_SIZE - off);
Jens Axboe912d35f2006-04-26 10:59:21 +02001469
1470 partial[buffers].offset = off;
1471 partial[buffers].len = plen;
1472
1473 off = 0;
1474 len -= plen;
1475 buffers++;
1476 }
1477
1478 /*
1479 * We didn't complete this iov, stop here since it probably
1480 * means we have to move some of this into a pipe to
1481 * be able to continue.
1482 */
1483 if (len)
1484 break;
1485
1486 /*
1487 * Don't continue if we mapped fewer pages than we asked for,
1488 * or if we mapped the max number of pages that we have
1489 * room for.
1490 */
Jens Axboe35f3d142010-05-20 10:43:18 +02001491 if (error < npages || buffers == pipe_buffers)
Jens Axboe912d35f2006-04-26 10:59:21 +02001492 break;
1493
1494 nr_vecs--;
1495 iov++;
1496 }
1497
Jens Axboe912d35f2006-04-26 10:59:21 +02001498 if (buffers)
1499 return buffers;
1500
1501 return error;
1502}
1503
Jens Axboe6a14b902007-06-14 13:08:55 +02001504static int pipe_to_user(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
1505 struct splice_desc *sd)
1506{
1507 char *src;
1508 int ret;
1509
Jens Axboecac36bb2007-06-14 13:10:48 +02001510 ret = buf->ops->confirm(pipe, buf);
Jens Axboe6a14b902007-06-14 13:08:55 +02001511 if (unlikely(ret))
1512 return ret;
1513
1514 /*
1515 * See if we can use the atomic maps, by prefaulting in the
1516 * pages and doing an atomic copy
1517 */
1518 if (!fault_in_pages_writeable(sd->u.userptr, sd->len)) {
1519 src = buf->ops->map(pipe, buf, 1);
1520 ret = __copy_to_user_inatomic(sd->u.userptr, src + buf->offset,
1521 sd->len);
1522 buf->ops->unmap(pipe, buf, src);
1523 if (!ret) {
1524 ret = sd->len;
1525 goto out;
1526 }
1527 }
1528
1529 /*
1530 * No dice, use slow non-atomic map and copy
1531 */
1532 src = buf->ops->map(pipe, buf, 0);
1533
1534 ret = sd->len;
1535 if (copy_to_user(sd->u.userptr, src + buf->offset, sd->len))
1536 ret = -EFAULT;
1537
Jens Axboe6866bef2007-10-16 10:01:29 +02001538 buf->ops->unmap(pipe, buf, src);
Jens Axboe6a14b902007-06-14 13:08:55 +02001539out:
1540 if (ret > 0)
1541 sd->u.userptr += ret;
Jens Axboe6a14b902007-06-14 13:08:55 +02001542 return ret;
1543}
1544
1545/*
1546 * For lack of a better implementation, implement vmsplice() to userspace
1547 * as a simple copy of the pipes pages to the user iov.
1548 */
1549static long vmsplice_to_user(struct file *file, const struct iovec __user *iov,
1550 unsigned long nr_segs, unsigned int flags)
1551{
1552 struct pipe_inode_info *pipe;
1553 struct splice_desc sd;
1554 ssize_t size;
1555 int error;
1556 long ret;
1557
1558 pipe = pipe_info(file->f_path.dentry->d_inode);
1559 if (!pipe)
1560 return -EBADF;
1561
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001562 pipe_lock(pipe);
Jens Axboe6a14b902007-06-14 13:08:55 +02001563
1564 error = ret = 0;
1565 while (nr_segs) {
1566 void __user *base;
1567 size_t len;
1568
1569 /*
1570 * Get user address base and length for this iovec.
1571 */
1572 error = get_user(base, &iov->iov_base);
1573 if (unlikely(error))
1574 break;
1575 error = get_user(len, &iov->iov_len);
1576 if (unlikely(error))
1577 break;
1578
1579 /*
1580 * Sanity check this iovec. 0 read succeeds.
1581 */
1582 if (unlikely(!len))
1583 break;
1584 if (unlikely(!base)) {
1585 error = -EFAULT;
1586 break;
1587 }
1588
Jens Axboe88119302008-02-08 08:49:14 -08001589 if (unlikely(!access_ok(VERIFY_WRITE, base, len))) {
1590 error = -EFAULT;
1591 break;
1592 }
1593
Jens Axboe6a14b902007-06-14 13:08:55 +02001594 sd.len = 0;
1595 sd.total_len = len;
1596 sd.flags = flags;
1597 sd.u.userptr = base;
1598 sd.pos = 0;
1599
1600 size = __splice_from_pipe(pipe, &sd, pipe_to_user);
1601 if (size < 0) {
1602 if (!ret)
1603 ret = size;
1604
1605 break;
1606 }
1607
1608 ret += size;
1609
1610 if (size < len)
1611 break;
1612
1613 nr_segs--;
1614 iov++;
1615 }
1616
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001617 pipe_unlock(pipe);
Jens Axboe6a14b902007-06-14 13:08:55 +02001618
1619 if (!ret)
1620 ret = error;
1621
1622 return ret;
1623}
1624
Jens Axboe912d35f2006-04-26 10:59:21 +02001625/*
1626 * vmsplice splices a user address range into a pipe. It can be thought of
1627 * as splice-from-memory, where the regular splice is splice-from-file (or
1628 * to file). In both cases the output is a pipe, naturally.
Jens Axboe912d35f2006-04-26 10:59:21 +02001629 */
Jens Axboe6a14b902007-06-14 13:08:55 +02001630static long vmsplice_to_pipe(struct file *file, const struct iovec __user *iov,
1631 unsigned long nr_segs, unsigned int flags)
Jens Axboe912d35f2006-04-26 10:59:21 +02001632{
Jens Axboeddac0d32006-11-04 12:49:32 +01001633 struct pipe_inode_info *pipe;
Jens Axboe35f3d142010-05-20 10:43:18 +02001634 struct page *pages[PIPE_DEF_BUFFERS];
1635 struct partial_page partial[PIPE_DEF_BUFFERS];
Jens Axboe912d35f2006-04-26 10:59:21 +02001636 struct splice_pipe_desc spd = {
1637 .pages = pages,
1638 .partial = partial,
1639 .flags = flags,
1640 .ops = &user_page_pipe_buf_ops,
Jens Axboebbdfc2f2007-11-06 23:29:47 -08001641 .spd_release = spd_release_page,
Jens Axboe912d35f2006-04-26 10:59:21 +02001642 };
Jens Axboe35f3d142010-05-20 10:43:18 +02001643 long ret;
Jens Axboe912d35f2006-04-26 10:59:21 +02001644
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001645 pipe = pipe_info(file->f_path.dentry->d_inode);
Jens Axboeddac0d32006-11-04 12:49:32 +01001646 if (!pipe)
Jens Axboe912d35f2006-04-26 10:59:21 +02001647 return -EBADF;
Jens Axboe912d35f2006-04-26 10:59:21 +02001648
Jens Axboe35f3d142010-05-20 10:43:18 +02001649 if (splice_grow_spd(pipe, &spd))
1650 return -ENOMEM;
Jens Axboe912d35f2006-04-26 10:59:21 +02001651
Jens Axboe35f3d142010-05-20 10:43:18 +02001652 spd.nr_pages = get_iovec_page_array(iov, nr_segs, spd.pages,
1653 spd.partial, flags & SPLICE_F_GIFT,
1654 pipe->buffers);
1655 if (spd.nr_pages <= 0)
1656 ret = spd.nr_pages;
1657 else
1658 ret = splice_to_pipe(pipe, &spd);
1659
1660 splice_shrink_spd(pipe, &spd);
1661 return ret;
Jens Axboe912d35f2006-04-26 10:59:21 +02001662}
1663
Jens Axboe6a14b902007-06-14 13:08:55 +02001664/*
1665 * Note that vmsplice only really supports true splicing _from_ user memory
1666 * to a pipe, not the other way around. Splicing from user memory is a simple
1667 * operation that can be supported without any funky alignment restrictions
1668 * or nasty vm tricks. We simply map in the user memory and fill them into
1669 * a pipe. The reverse isn't quite as easy, though. There are two possible
1670 * solutions for that:
1671 *
1672 * - memcpy() the data internally, at which point we might as well just
1673 * do a regular read() on the buffer anyway.
1674 * - Lots of nasty vm tricks, that are neither fast nor flexible (it
1675 * has restriction limitations on both ends of the pipe).
1676 *
1677 * Currently we punt and implement it as a normal copy, see pipe_to_user().
1678 *
1679 */
Heiko Carstens836f92a2009-01-14 14:14:33 +01001680SYSCALL_DEFINE4(vmsplice, int, fd, const struct iovec __user *, iov,
1681 unsigned long, nr_segs, unsigned int, flags)
Jens Axboe912d35f2006-04-26 10:59:21 +02001682{
1683 struct file *file;
1684 long error;
1685 int fput;
1686
Jens Axboe6a14b902007-06-14 13:08:55 +02001687 if (unlikely(nr_segs > UIO_MAXIOV))
1688 return -EINVAL;
1689 else if (unlikely(!nr_segs))
1690 return 0;
1691
Jens Axboe912d35f2006-04-26 10:59:21 +02001692 error = -EBADF;
1693 file = fget_light(fd, &fput);
1694 if (file) {
1695 if (file->f_mode & FMODE_WRITE)
Jens Axboe6a14b902007-06-14 13:08:55 +02001696 error = vmsplice_to_pipe(file, iov, nr_segs, flags);
1697 else if (file->f_mode & FMODE_READ)
1698 error = vmsplice_to_user(file, iov, nr_segs, flags);
Jens Axboe912d35f2006-04-26 10:59:21 +02001699
1700 fput_light(file, fput);
1701 }
1702
1703 return error;
1704}
1705
Heiko Carstens836f92a2009-01-14 14:14:33 +01001706SYSCALL_DEFINE6(splice, int, fd_in, loff_t __user *, off_in,
1707 int, fd_out, loff_t __user *, off_out,
1708 size_t, len, unsigned int, flags)
Jens Axboe5274f052006-03-30 15:15:30 +02001709{
1710 long error;
1711 struct file *in, *out;
1712 int fput_in, fput_out;
1713
1714 if (unlikely(!len))
1715 return 0;
1716
1717 error = -EBADF;
Ingo Molnar529565d2006-04-10 15:18:58 +02001718 in = fget_light(fd_in, &fput_in);
Jens Axboe5274f052006-03-30 15:15:30 +02001719 if (in) {
1720 if (in->f_mode & FMODE_READ) {
Ingo Molnar529565d2006-04-10 15:18:58 +02001721 out = fget_light(fd_out, &fput_out);
Jens Axboe5274f052006-03-30 15:15:30 +02001722 if (out) {
1723 if (out->f_mode & FMODE_WRITE)
Ingo Molnar529565d2006-04-10 15:18:58 +02001724 error = do_splice(in, off_in,
1725 out, off_out,
1726 len, flags);
Jens Axboe5274f052006-03-30 15:15:30 +02001727 fput_light(out, fput_out);
1728 }
1729 }
1730
1731 fput_light(in, fput_in);
1732 }
1733
1734 return error;
1735}
Jens Axboe70524492006-04-11 15:51:17 +02001736
1737/*
Jens Axboeaadd06e2006-07-10 11:00:01 +02001738 * Make sure there's data to read. Wait for input if we can, otherwise
1739 * return an appropriate error.
1740 */
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001741static int ipipe_prep(struct pipe_inode_info *pipe, unsigned int flags)
Jens Axboeaadd06e2006-07-10 11:00:01 +02001742{
1743 int ret;
1744
1745 /*
1746 * Check ->nrbufs without the inode lock first. This function
1747 * is speculative anyways, so missing one is ok.
1748 */
1749 if (pipe->nrbufs)
1750 return 0;
1751
1752 ret = 0;
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001753 pipe_lock(pipe);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001754
1755 while (!pipe->nrbufs) {
1756 if (signal_pending(current)) {
1757 ret = -ERESTARTSYS;
1758 break;
1759 }
1760 if (!pipe->writers)
1761 break;
1762 if (!pipe->waiting_writers) {
1763 if (flags & SPLICE_F_NONBLOCK) {
1764 ret = -EAGAIN;
1765 break;
1766 }
1767 }
1768 pipe_wait(pipe);
1769 }
1770
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001771 pipe_unlock(pipe);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001772 return ret;
1773}
1774
1775/*
1776 * Make sure there's writeable room. Wait for room if we can, otherwise
1777 * return an appropriate error.
1778 */
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001779static int opipe_prep(struct pipe_inode_info *pipe, unsigned int flags)
Jens Axboeaadd06e2006-07-10 11:00:01 +02001780{
1781 int ret;
1782
1783 /*
1784 * Check ->nrbufs without the inode lock first. This function
1785 * is speculative anyways, so missing one is ok.
1786 */
Jens Axboe35f3d142010-05-20 10:43:18 +02001787 if (pipe->nrbufs < pipe->buffers)
Jens Axboeaadd06e2006-07-10 11:00:01 +02001788 return 0;
1789
1790 ret = 0;
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001791 pipe_lock(pipe);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001792
Jens Axboe35f3d142010-05-20 10:43:18 +02001793 while (pipe->nrbufs >= pipe->buffers) {
Jens Axboeaadd06e2006-07-10 11:00:01 +02001794 if (!pipe->readers) {
1795 send_sig(SIGPIPE, current, 0);
1796 ret = -EPIPE;
1797 break;
1798 }
1799 if (flags & SPLICE_F_NONBLOCK) {
1800 ret = -EAGAIN;
1801 break;
1802 }
1803 if (signal_pending(current)) {
1804 ret = -ERESTARTSYS;
1805 break;
1806 }
1807 pipe->waiting_writers++;
1808 pipe_wait(pipe);
1809 pipe->waiting_writers--;
1810 }
1811
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001812 pipe_unlock(pipe);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001813 return ret;
1814}
1815
1816/*
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001817 * Splice contents of ipipe to opipe.
1818 */
1819static int splice_pipe_to_pipe(struct pipe_inode_info *ipipe,
1820 struct pipe_inode_info *opipe,
1821 size_t len, unsigned int flags)
1822{
1823 struct pipe_buffer *ibuf, *obuf;
1824 int ret = 0, nbuf;
1825 bool input_wakeup = false;
1826
1827
1828retry:
1829 ret = ipipe_prep(ipipe, flags);
1830 if (ret)
1831 return ret;
1832
1833 ret = opipe_prep(opipe, flags);
1834 if (ret)
1835 return ret;
1836
1837 /*
1838 * Potential ABBA deadlock, work around it by ordering lock
1839 * grabbing by pipe info address. Otherwise two different processes
1840 * could deadlock (one doing tee from A -> B, the other from B -> A).
1841 */
1842 pipe_double_lock(ipipe, opipe);
1843
1844 do {
1845 if (!opipe->readers) {
1846 send_sig(SIGPIPE, current, 0);
1847 if (!ret)
1848 ret = -EPIPE;
1849 break;
1850 }
1851
1852 if (!ipipe->nrbufs && !ipipe->writers)
1853 break;
1854
1855 /*
1856 * Cannot make any progress, because either the input
1857 * pipe is empty or the output pipe is full.
1858 */
Jens Axboe35f3d142010-05-20 10:43:18 +02001859 if (!ipipe->nrbufs || opipe->nrbufs >= opipe->buffers) {
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001860 /* Already processed some buffers, break */
1861 if (ret)
1862 break;
1863
1864 if (flags & SPLICE_F_NONBLOCK) {
1865 ret = -EAGAIN;
1866 break;
1867 }
1868
1869 /*
1870 * We raced with another reader/writer and haven't
1871 * managed to process any buffers. A zero return
1872 * value means EOF, so retry instead.
1873 */
1874 pipe_unlock(ipipe);
1875 pipe_unlock(opipe);
1876 goto retry;
1877 }
1878
1879 ibuf = ipipe->bufs + ipipe->curbuf;
Jens Axboe35f3d142010-05-20 10:43:18 +02001880 nbuf = (opipe->curbuf + opipe->nrbufs) & (opipe->buffers - 1);
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001881 obuf = opipe->bufs + nbuf;
1882
1883 if (len >= ibuf->len) {
1884 /*
1885 * Simply move the whole buffer from ipipe to opipe
1886 */
1887 *obuf = *ibuf;
1888 ibuf->ops = NULL;
1889 opipe->nrbufs++;
Jens Axboe35f3d142010-05-20 10:43:18 +02001890 ipipe->curbuf = (ipipe->curbuf + 1) & (ipipe->buffers - 1);
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001891 ipipe->nrbufs--;
1892 input_wakeup = true;
1893 } else {
1894 /*
1895 * Get a reference to this pipe buffer,
1896 * so we can copy the contents over.
1897 */
1898 ibuf->ops->get(ipipe, ibuf);
1899 *obuf = *ibuf;
1900
1901 /*
1902 * Don't inherit the gift flag, we need to
1903 * prevent multiple steals of this page.
1904 */
1905 obuf->flags &= ~PIPE_BUF_FLAG_GIFT;
1906
1907 obuf->len = len;
1908 opipe->nrbufs++;
1909 ibuf->offset += obuf->len;
1910 ibuf->len -= obuf->len;
1911 }
1912 ret += obuf->len;
1913 len -= obuf->len;
1914 } while (len);
1915
1916 pipe_unlock(ipipe);
1917 pipe_unlock(opipe);
1918
1919 /*
1920 * If we put data in the output pipe, wakeup any potential readers.
1921 */
1922 if (ret > 0) {
1923 smp_mb();
1924 if (waitqueue_active(&opipe->wait))
1925 wake_up_interruptible(&opipe->wait);
1926 kill_fasync(&opipe->fasync_readers, SIGIO, POLL_IN);
1927 }
1928 if (input_wakeup)
1929 wakeup_pipe_writers(ipipe);
1930
1931 return ret;
1932}
1933
1934/*
Jens Axboe70524492006-04-11 15:51:17 +02001935 * Link contents of ipipe to opipe.
1936 */
1937static int link_pipe(struct pipe_inode_info *ipipe,
1938 struct pipe_inode_info *opipe,
1939 size_t len, unsigned int flags)
1940{
1941 struct pipe_buffer *ibuf, *obuf;
Jens Axboeaadd06e2006-07-10 11:00:01 +02001942 int ret = 0, i = 0, nbuf;
Jens Axboe70524492006-04-11 15:51:17 +02001943
1944 /*
1945 * Potential ABBA deadlock, work around it by ordering lock
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001946 * grabbing by pipe info address. Otherwise two different processes
Jens Axboe70524492006-04-11 15:51:17 +02001947 * could deadlock (one doing tee from A -> B, the other from B -> A).
1948 */
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001949 pipe_double_lock(ipipe, opipe);
Jens Axboe70524492006-04-11 15:51:17 +02001950
Jens Axboeaadd06e2006-07-10 11:00:01 +02001951 do {
Jens Axboe70524492006-04-11 15:51:17 +02001952 if (!opipe->readers) {
1953 send_sig(SIGPIPE, current, 0);
1954 if (!ret)
1955 ret = -EPIPE;
1956 break;
1957 }
Jens Axboe70524492006-04-11 15:51:17 +02001958
1959 /*
Jens Axboeaadd06e2006-07-10 11:00:01 +02001960 * If we have iterated all input buffers or ran out of
1961 * output room, break.
Jens Axboe70524492006-04-11 15:51:17 +02001962 */
Jens Axboe35f3d142010-05-20 10:43:18 +02001963 if (i >= ipipe->nrbufs || opipe->nrbufs >= opipe->buffers)
Jens Axboe70524492006-04-11 15:51:17 +02001964 break;
Jens Axboeaadd06e2006-07-10 11:00:01 +02001965
Jens Axboe35f3d142010-05-20 10:43:18 +02001966 ibuf = ipipe->bufs + ((ipipe->curbuf + i) & (ipipe->buffers-1));
1967 nbuf = (opipe->curbuf + opipe->nrbufs) & (opipe->buffers - 1);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001968
Jens Axboe2a272502006-04-19 15:56:40 +02001969 /*
Jens Axboeaadd06e2006-07-10 11:00:01 +02001970 * Get a reference to this pipe buffer,
1971 * so we can copy the contents over.
Jens Axboe2a272502006-04-19 15:56:40 +02001972 */
Jens Axboeaadd06e2006-07-10 11:00:01 +02001973 ibuf->ops->get(ipipe, ibuf);
Jens Axboe70524492006-04-11 15:51:17 +02001974
Jens Axboeaadd06e2006-07-10 11:00:01 +02001975 obuf = opipe->bufs + nbuf;
1976 *obuf = *ibuf;
Jens Axboe70524492006-04-11 15:51:17 +02001977
Jens Axboeaadd06e2006-07-10 11:00:01 +02001978 /*
1979 * Don't inherit the gift flag, we need to
1980 * prevent multiple steals of this page.
1981 */
1982 obuf->flags &= ~PIPE_BUF_FLAG_GIFT;
1983
1984 if (obuf->len > len)
1985 obuf->len = len;
1986
1987 opipe->nrbufs++;
1988 ret += obuf->len;
1989 len -= obuf->len;
1990 i++;
1991 } while (len);
Jens Axboe70524492006-04-11 15:51:17 +02001992
Jens Axboe02cf01a2008-02-20 10:34:51 +01001993 /*
1994 * return EAGAIN if we have the potential of some data in the
1995 * future, otherwise just return 0
1996 */
1997 if (!ret && ipipe->waiting_writers && (flags & SPLICE_F_NONBLOCK))
1998 ret = -EAGAIN;
1999
Miklos Szeredi61e0d472009-04-14 19:48:41 +02002000 pipe_unlock(ipipe);
2001 pipe_unlock(opipe);
Jens Axboe70524492006-04-11 15:51:17 +02002002
Jens Axboeaadd06e2006-07-10 11:00:01 +02002003 /*
2004 * If we put data in the output pipe, wakeup any potential readers.
2005 */
2006 if (ret > 0) {
Jens Axboe70524492006-04-11 15:51:17 +02002007 smp_mb();
2008 if (waitqueue_active(&opipe->wait))
2009 wake_up_interruptible(&opipe->wait);
2010 kill_fasync(&opipe->fasync_readers, SIGIO, POLL_IN);
2011 }
2012
2013 return ret;
2014}
2015
2016/*
2017 * This is a tee(1) implementation that works on pipes. It doesn't copy
2018 * any data, it simply references the 'in' pages on the 'out' pipe.
2019 * The 'flags' used are the SPLICE_F_* variants, currently the only
2020 * applicable one is SPLICE_F_NONBLOCK.
2021 */
2022static long do_tee(struct file *in, struct file *out, size_t len,
2023 unsigned int flags)
2024{
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08002025 struct pipe_inode_info *ipipe = pipe_info(in->f_path.dentry->d_inode);
2026 struct pipe_inode_info *opipe = pipe_info(out->f_path.dentry->d_inode);
Jens Axboeaadd06e2006-07-10 11:00:01 +02002027 int ret = -EINVAL;
Jens Axboe70524492006-04-11 15:51:17 +02002028
2029 /*
Jens Axboeaadd06e2006-07-10 11:00:01 +02002030 * Duplicate the contents of ipipe to opipe without actually
2031 * copying the data.
Jens Axboe70524492006-04-11 15:51:17 +02002032 */
Jens Axboeaadd06e2006-07-10 11:00:01 +02002033 if (ipipe && opipe && ipipe != opipe) {
2034 /*
2035 * Keep going, unless we encounter an error. The ipipe/opipe
2036 * ordering doesn't really matter.
2037 */
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02002038 ret = ipipe_prep(ipipe, flags);
Jens Axboeaadd06e2006-07-10 11:00:01 +02002039 if (!ret) {
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02002040 ret = opipe_prep(opipe, flags);
Jens Axboe02cf01a2008-02-20 10:34:51 +01002041 if (!ret)
Jens Axboeaadd06e2006-07-10 11:00:01 +02002042 ret = link_pipe(ipipe, opipe, len, flags);
Jens Axboeaadd06e2006-07-10 11:00:01 +02002043 }
2044 }
Jens Axboe70524492006-04-11 15:51:17 +02002045
Jens Axboeaadd06e2006-07-10 11:00:01 +02002046 return ret;
Jens Axboe70524492006-04-11 15:51:17 +02002047}
2048
Heiko Carstens836f92a2009-01-14 14:14:33 +01002049SYSCALL_DEFINE4(tee, int, fdin, int, fdout, size_t, len, unsigned int, flags)
Jens Axboe70524492006-04-11 15:51:17 +02002050{
2051 struct file *in;
2052 int error, fput_in;
2053
2054 if (unlikely(!len))
2055 return 0;
2056
2057 error = -EBADF;
2058 in = fget_light(fdin, &fput_in);
2059 if (in) {
2060 if (in->f_mode & FMODE_READ) {
2061 int fput_out;
2062 struct file *out = fget_light(fdout, &fput_out);
2063
2064 if (out) {
2065 if (out->f_mode & FMODE_WRITE)
2066 error = do_tee(in, out, len, flags);
2067 fput_light(out, fput_out);
2068 }
2069 }
2070 fput_light(in, fput_in);
2071 }
2072
2073 return error;
2074}