blob: eefd96b1d7fbfdc7c5873ca45e551efffcf548e1 [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>
Jens Axboe5274f052006-03-30 15:15:30 +020033
Jens Axboe83f91352006-04-02 23:05:09 +020034/*
35 * Attempt to steal a page from a pipe buffer. This should perhaps go into
36 * a vm helper function, it's already simplified quite a bit by the
37 * addition of remove_mapping(). If success is returned, the caller may
38 * attempt to reuse this page for another destination.
39 */
Jens Axboe76ad4d12006-05-03 10:41:33 +020040static int page_cache_pipe_buf_steal(struct pipe_inode_info *pipe,
Jens Axboe5abc97a2006-03-30 15:16:46 +020041 struct pipe_buffer *buf)
42{
43 struct page *page = buf->page;
Jens Axboe9e94cd42006-06-20 15:01:12 +020044 struct address_space *mapping;
Jens Axboe5abc97a2006-03-30 15:16:46 +020045
Jens Axboe9e0267c2006-04-19 15:57:31 +020046 lock_page(page);
47
Jens Axboe9e94cd42006-06-20 15:01:12 +020048 mapping = page_mapping(page);
49 if (mapping) {
50 WARN_ON(!PageUptodate(page));
Jens Axboe5abc97a2006-03-30 15:16:46 +020051
Jens Axboe9e94cd42006-06-20 15:01:12 +020052 /*
53 * At least for ext2 with nobh option, we need to wait on
54 * writeback completing on this page, since we'll remove it
55 * from the pagecache. Otherwise truncate wont wait on the
56 * page, allowing the disk blocks to be reused by someone else
57 * before we actually wrote our data to them. fs corruption
58 * ensues.
59 */
60 wait_on_page_writeback(page);
Jens Axboead8d6f02006-04-02 23:10:32 +020061
David Howells266cf652009-04-03 16:42:36 +010062 if (page_has_private(page) &&
63 !try_to_release_page(page, GFP_KERNEL))
Jens Axboeca39d652008-05-20 21:27:41 +020064 goto out_unlock;
Jens Axboe4f6f0bd2006-04-02 23:04:46 +020065
Jens Axboe9e94cd42006-06-20 15:01:12 +020066 /*
67 * If we succeeded in removing the mapping, set LRU flag
68 * and return good.
69 */
70 if (remove_mapping(mapping, page)) {
71 buf->flags |= PIPE_BUF_FLAG_LRU;
72 return 0;
73 }
Jens Axboe9e0267c2006-04-19 15:57:31 +020074 }
Jens Axboe5abc97a2006-03-30 15:16:46 +020075
Jens Axboe9e94cd42006-06-20 15:01:12 +020076 /*
77 * Raced with truncate or failed to remove page from current
78 * address space, unlock and return failure.
79 */
Jens Axboeca39d652008-05-20 21:27:41 +020080out_unlock:
Jens Axboe9e94cd42006-06-20 15:01:12 +020081 unlock_page(page);
82 return 1;
Jens Axboe5abc97a2006-03-30 15:16:46 +020083}
84
Jens Axboe76ad4d12006-05-03 10:41:33 +020085static void page_cache_pipe_buf_release(struct pipe_inode_info *pipe,
Jens Axboe5274f052006-03-30 15:15:30 +020086 struct pipe_buffer *buf)
87{
88 page_cache_release(buf->page);
Jens Axboe14328732006-05-03 10:35:26 +020089 buf->flags &= ~PIPE_BUF_FLAG_LRU;
Jens Axboe5274f052006-03-30 15:15:30 +020090}
91
Jens Axboe08457182007-06-12 20:51:32 +020092/*
93 * Check whether the contents of buf is OK to access. Since the content
94 * is a page cache page, IO may be in flight.
95 */
Jens Axboecac36bb2007-06-14 13:10:48 +020096static int page_cache_pipe_buf_confirm(struct pipe_inode_info *pipe,
97 struct pipe_buffer *buf)
Jens Axboe5274f052006-03-30 15:15:30 +020098{
99 struct page *page = buf->page;
Jens Axboe49d0b212006-04-10 09:04:41 +0200100 int err;
Jens Axboe5274f052006-03-30 15:15:30 +0200101
102 if (!PageUptodate(page)) {
Jens Axboe49d0b212006-04-10 09:04:41 +0200103 lock_page(page);
104
105 /*
106 * Page got truncated/unhashed. This will cause a 0-byte
Ingo Molnar73d62d82006-04-11 13:57:21 +0200107 * splice, if this is the first page.
Jens Axboe49d0b212006-04-10 09:04:41 +0200108 */
109 if (!page->mapping) {
110 err = -ENODATA;
111 goto error;
112 }
113
114 /*
Ingo Molnar73d62d82006-04-11 13:57:21 +0200115 * Uh oh, read-error from disk.
Jens Axboe49d0b212006-04-10 09:04:41 +0200116 */
117 if (!PageUptodate(page)) {
118 err = -EIO;
119 goto error;
120 }
121
122 /*
Jens Axboef84d7512006-05-01 19:59:03 +0200123 * Page is ok afterall, we are done.
Jens Axboe49d0b212006-04-10 09:04:41 +0200124 */
Jens Axboe5274f052006-03-30 15:15:30 +0200125 unlock_page(page);
Jens Axboe5274f052006-03-30 15:15:30 +0200126 }
127
Jens Axboef84d7512006-05-01 19:59:03 +0200128 return 0;
Jens Axboe49d0b212006-04-10 09:04:41 +0200129error:
130 unlock_page(page);
Jens Axboef84d7512006-05-01 19:59:03 +0200131 return err;
Jens Axboe70524492006-04-11 15:51:17 +0200132}
133
Eric Dumazetd4c3cca2006-12-13 00:34:04 -0800134static const struct pipe_buf_operations page_cache_pipe_buf_ops = {
Jens Axboe5274f052006-03-30 15:15:30 +0200135 .can_merge = 0,
Jens Axboef84d7512006-05-01 19:59:03 +0200136 .map = generic_pipe_buf_map,
137 .unmap = generic_pipe_buf_unmap,
Jens Axboecac36bb2007-06-14 13:10:48 +0200138 .confirm = page_cache_pipe_buf_confirm,
Jens Axboe5274f052006-03-30 15:15:30 +0200139 .release = page_cache_pipe_buf_release,
Jens Axboe5abc97a2006-03-30 15:16:46 +0200140 .steal = page_cache_pipe_buf_steal,
Jens Axboef84d7512006-05-01 19:59:03 +0200141 .get = generic_pipe_buf_get,
Jens Axboe5274f052006-03-30 15:15:30 +0200142};
143
Jens Axboe912d35f2006-04-26 10:59:21 +0200144static int user_page_pipe_buf_steal(struct pipe_inode_info *pipe,
145 struct pipe_buffer *buf)
146{
Jens Axboe7afa6fd2006-05-01 20:02:33 +0200147 if (!(buf->flags & PIPE_BUF_FLAG_GIFT))
148 return 1;
149
Jens Axboe14328732006-05-03 10:35:26 +0200150 buf->flags |= PIPE_BUF_FLAG_LRU;
Jens Axboe330ab712006-05-02 15:29:57 +0200151 return generic_pipe_buf_steal(pipe, buf);
Jens Axboe912d35f2006-04-26 10:59:21 +0200152}
153
Eric Dumazetd4c3cca2006-12-13 00:34:04 -0800154static const struct pipe_buf_operations user_page_pipe_buf_ops = {
Jens Axboe912d35f2006-04-26 10:59:21 +0200155 .can_merge = 0,
Jens Axboef84d7512006-05-01 19:59:03 +0200156 .map = generic_pipe_buf_map,
157 .unmap = generic_pipe_buf_unmap,
Jens Axboecac36bb2007-06-14 13:10:48 +0200158 .confirm = generic_pipe_buf_confirm,
Jens Axboe912d35f2006-04-26 10:59:21 +0200159 .release = page_cache_pipe_buf_release,
160 .steal = user_page_pipe_buf_steal,
Jens Axboef84d7512006-05-01 19:59:03 +0200161 .get = generic_pipe_buf_get,
Jens Axboe912d35f2006-04-26 10:59:21 +0200162};
163
Jens Axboe932cc6d2007-06-21 13:10:21 +0200164/**
165 * splice_to_pipe - fill passed data into a pipe
166 * @pipe: pipe to fill
167 * @spd: data to fill
168 *
169 * Description:
Randy Dunlap79685b82007-07-27 08:08:51 +0200170 * @spd contains a map of pages and len/offset tuples, along with
Jens Axboe932cc6d2007-06-21 13:10:21 +0200171 * the struct pipe_buf_operations associated with these pages. This
172 * function will link that data to the pipe.
173 *
Jens Axboe83f91352006-04-02 23:05:09 +0200174 */
Jens Axboed6b29d72007-06-04 09:59:47 +0200175ssize_t splice_to_pipe(struct pipe_inode_info *pipe,
176 struct splice_pipe_desc *spd)
Jens Axboe5274f052006-03-30 15:15:30 +0200177{
Jens Axboe00de00b2007-06-15 13:14:22 +0200178 unsigned int spd_pages = spd->nr_pages;
Jens Axboe912d35f2006-04-26 10:59:21 +0200179 int ret, do_wakeup, page_nr;
Jens Axboe5274f052006-03-30 15:15:30 +0200180
181 ret = 0;
182 do_wakeup = 0;
Jens Axboe912d35f2006-04-26 10:59:21 +0200183 page_nr = 0;
Jens Axboe5274f052006-03-30 15:15:30 +0200184
Miklos Szeredi61e0d472009-04-14 19:48:41 +0200185 pipe_lock(pipe);
Jens Axboe5274f052006-03-30 15:15:30 +0200186
Jens Axboe5274f052006-03-30 15:15:30 +0200187 for (;;) {
Ingo Molnar3a326a22006-04-10 15:18:35 +0200188 if (!pipe->readers) {
Jens Axboe5274f052006-03-30 15:15:30 +0200189 send_sig(SIGPIPE, current, 0);
190 if (!ret)
191 ret = -EPIPE;
192 break;
193 }
194
Jens Axboe6f767b02006-04-11 13:53:56 +0200195 if (pipe->nrbufs < PIPE_BUFFERS) {
196 int newbuf = (pipe->curbuf + pipe->nrbufs) & (PIPE_BUFFERS - 1);
Ingo Molnar3a326a22006-04-10 15:18:35 +0200197 struct pipe_buffer *buf = pipe->bufs + newbuf;
Jens Axboe5274f052006-03-30 15:15:30 +0200198
Jens Axboe912d35f2006-04-26 10:59:21 +0200199 buf->page = spd->pages[page_nr];
200 buf->offset = spd->partial[page_nr].offset;
201 buf->len = spd->partial[page_nr].len;
Jens Axboe497f9622007-06-11 12:00:45 +0200202 buf->private = spd->partial[page_nr].private;
Jens Axboe912d35f2006-04-26 10:59:21 +0200203 buf->ops = spd->ops;
Jens Axboe7afa6fd2006-05-01 20:02:33 +0200204 if (spd->flags & SPLICE_F_GIFT)
205 buf->flags |= PIPE_BUF_FLAG_GIFT;
206
Jens Axboe6f767b02006-04-11 13:53:56 +0200207 pipe->nrbufs++;
Jens Axboe912d35f2006-04-26 10:59:21 +0200208 page_nr++;
209 ret += buf->len;
210
Jens Axboe6f767b02006-04-11 13:53:56 +0200211 if (pipe->inode)
212 do_wakeup = 1;
Jens Axboe5274f052006-03-30 15:15:30 +0200213
Jens Axboe912d35f2006-04-26 10:59:21 +0200214 if (!--spd->nr_pages)
Jens Axboe5274f052006-03-30 15:15:30 +0200215 break;
Jens Axboe6f767b02006-04-11 13:53:56 +0200216 if (pipe->nrbufs < PIPE_BUFFERS)
Jens Axboe5274f052006-03-30 15:15:30 +0200217 continue;
218
219 break;
220 }
221
Jens Axboe912d35f2006-04-26 10:59:21 +0200222 if (spd->flags & SPLICE_F_NONBLOCK) {
Linus Torvalds29e35092006-04-02 12:46:35 -0700223 if (!ret)
224 ret = -EAGAIN;
225 break;
226 }
227
Jens Axboe5274f052006-03-30 15:15:30 +0200228 if (signal_pending(current)) {
229 if (!ret)
230 ret = -ERESTARTSYS;
231 break;
232 }
233
234 if (do_wakeup) {
Jens Axboec0bd1f62006-04-10 09:03:32 +0200235 smp_mb();
Ingo Molnar3a326a22006-04-10 15:18:35 +0200236 if (waitqueue_active(&pipe->wait))
237 wake_up_interruptible_sync(&pipe->wait);
238 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
Jens Axboe5274f052006-03-30 15:15:30 +0200239 do_wakeup = 0;
240 }
241
Ingo Molnar3a326a22006-04-10 15:18:35 +0200242 pipe->waiting_writers++;
243 pipe_wait(pipe);
244 pipe->waiting_writers--;
Jens Axboe5274f052006-03-30 15:15:30 +0200245 }
246
Miklos Szeredi61e0d472009-04-14 19:48:41 +0200247 pipe_unlock(pipe);
Jens Axboe5274f052006-03-30 15:15:30 +0200248
Miklos Szeredi61e0d472009-04-14 19:48:41 +0200249 if (do_wakeup) {
250 smp_mb();
251 if (waitqueue_active(&pipe->wait))
252 wake_up_interruptible(&pipe->wait);
253 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
Jens Axboe5274f052006-03-30 15:15:30 +0200254 }
255
Jens Axboe00de00b2007-06-15 13:14:22 +0200256 while (page_nr < spd_pages)
Jens Axboebbdfc2f2007-11-06 23:29:47 -0800257 spd->spd_release(spd, page_nr++);
Jens Axboe5274f052006-03-30 15:15:30 +0200258
259 return ret;
260}
261
Jens Axboebbdfc2f2007-11-06 23:29:47 -0800262static void spd_release_page(struct splice_pipe_desc *spd, unsigned int i)
263{
264 page_cache_release(spd->pages[i]);
265}
266
Ingo Molnar3a326a22006-04-10 15:18:35 +0200267static int
Jens Axboecbb7e572006-04-11 14:57:50 +0200268__generic_file_splice_read(struct file *in, loff_t *ppos,
269 struct pipe_inode_info *pipe, size_t len,
270 unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +0200271{
272 struct address_space *mapping = in->f_mapping;
Fengguang Wud8983912007-07-19 01:48:06 -0700273 unsigned int loff, nr_pages, req_pages;
Jens Axboe16c523d2006-04-10 09:03:58 +0200274 struct page *pages[PIPE_BUFFERS];
Jens Axboe912d35f2006-04-26 10:59:21 +0200275 struct partial_page partial[PIPE_BUFFERS];
Jens Axboe5274f052006-03-30 15:15:30 +0200276 struct page *page;
Jens Axboe91ad66e2006-04-19 15:55:10 +0200277 pgoff_t index, end_index;
278 loff_t isize;
Jens Axboeeb207962006-04-27 11:05:22 +0200279 int error, page_nr;
Jens Axboe912d35f2006-04-26 10:59:21 +0200280 struct splice_pipe_desc spd = {
281 .pages = pages,
282 .partial = partial,
283 .flags = flags,
284 .ops = &page_cache_pipe_buf_ops,
Jens Axboebbdfc2f2007-11-06 23:29:47 -0800285 .spd_release = spd_release_page,
Jens Axboe912d35f2006-04-26 10:59:21 +0200286 };
Jens Axboe5274f052006-03-30 15:15:30 +0200287
Jens Axboecbb7e572006-04-11 14:57:50 +0200288 index = *ppos >> PAGE_CACHE_SHIFT;
Jens Axboe912d35f2006-04-26 10:59:21 +0200289 loff = *ppos & ~PAGE_CACHE_MASK;
Fengguang Wud8983912007-07-19 01:48:06 -0700290 req_pages = (len + loff + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
291 nr_pages = min(req_pages, (unsigned)PIPE_BUFFERS);
Jens Axboe5274f052006-03-30 15:15:30 +0200292
293 /*
Jens Axboeeb207962006-04-27 11:05:22 +0200294 * Lookup the (hopefully) full range of pages we need.
295 */
296 spd.nr_pages = find_get_pages_contig(mapping, index, nr_pages, pages);
Fengguang Wu431a4822007-07-19 01:48:05 -0700297 index += spd.nr_pages;
Jens Axboeeb207962006-04-27 11:05:22 +0200298
299 /*
300 * If find_get_pages_contig() returned fewer pages than we needed,
Fengguang Wu431a4822007-07-19 01:48:05 -0700301 * readahead/allocate the rest and fill in the holes.
Jens Axboeeb207962006-04-27 11:05:22 +0200302 */
Fengguang Wu431a4822007-07-19 01:48:05 -0700303 if (spd.nr_pages < nr_pages)
Rusty Russellcf914a72007-07-19 01:48:08 -0700304 page_cache_sync_readahead(mapping, &in->f_ra, in,
305 index, req_pages - spd.nr_pages);
Fengguang Wu431a4822007-07-19 01:48:05 -0700306
Jens Axboe932cc6d2007-06-21 13:10:21 +0200307 error = 0;
Jens Axboeeb207962006-04-27 11:05:22 +0200308 while (spd.nr_pages < nr_pages) {
309 /*
310 * Page could be there, find_get_pages_contig() breaks on
311 * the first hole.
312 */
313 page = find_get_page(mapping, index);
314 if (!page) {
Jens Axboee27dedd2006-05-01 19:59:54 +0200315 /*
Jens Axboeeb207962006-04-27 11:05:22 +0200316 * page didn't exist, allocate one.
317 */
318 page = page_cache_alloc_cold(mapping);
319 if (!page)
320 break;
321
322 error = add_to_page_cache_lru(page, mapping, index,
Hugh Dickins4cd13502008-04-03 23:35:22 +0100323 mapping_gfp_mask(mapping));
Jens Axboeeb207962006-04-27 11:05:22 +0200324 if (unlikely(error)) {
325 page_cache_release(page);
Jens Axboea0548872006-05-03 10:58:22 +0200326 if (error == -EEXIST)
327 continue;
Jens Axboeeb207962006-04-27 11:05:22 +0200328 break;
329 }
330 /*
331 * add_to_page_cache() locks the page, unlock it
332 * to avoid convoluting the logic below even more.
333 */
334 unlock_page(page);
335 }
336
337 pages[spd.nr_pages++] = page;
338 index++;
339 }
340
341 /*
342 * Now loop over the map and see if we need to start IO on any
343 * pages, fill in the partial map, etc.
344 */
345 index = *ppos >> PAGE_CACHE_SHIFT;
346 nr_pages = spd.nr_pages;
347 spd.nr_pages = 0;
348 for (page_nr = 0; page_nr < nr_pages; page_nr++) {
Jens Axboe82aa5d62006-04-20 13:05:48 +0200349 unsigned int this_len;
350
351 if (!len)
352 break;
353
354 /*
355 * this_len is the max we'll use from this page
356 */
Andrew Mortonba5f5d92006-04-25 15:33:34 +0200357 this_len = min_t(unsigned long, len, PAGE_CACHE_SIZE - loff);
Jens Axboeeb207962006-04-27 11:05:22 +0200358 page = pages[page_nr];
Jens Axboe7480a902006-04-11 13:52:47 +0200359
Fengguang Wua08a1662007-07-19 01:48:03 -0700360 if (PageReadahead(page))
Rusty Russellcf914a72007-07-19 01:48:08 -0700361 page_cache_async_readahead(mapping, &in->f_ra, in,
Fengguang Wud8983912007-07-19 01:48:06 -0700362 page, index, req_pages - page_nr);
Fengguang Wua08a1662007-07-19 01:48:03 -0700363
Jens Axboe7480a902006-04-11 13:52:47 +0200364 /*
365 * If the page isn't uptodate, we may need to start io on it
366 */
367 if (!PageUptodate(page)) {
Jens Axboec4f895c2006-04-19 15:56:12 +0200368 /*
369 * If in nonblock mode then dont block on waiting
370 * for an in-flight io page
371 */
Fengguang Wu9ae9d682007-05-08 08:44:36 +0200372 if (flags & SPLICE_F_NONBLOCK) {
Nick Piggin529ae9a2008-08-02 12:01:03 +0200373 if (!trylock_page(page)) {
Jens Axboe8191ecd2008-04-10 08:24:25 +0200374 error = -EAGAIN;
Fengguang Wu9ae9d682007-05-08 08:44:36 +0200375 break;
Jens Axboe8191ecd2008-04-10 08:24:25 +0200376 }
Fengguang Wu9ae9d682007-05-08 08:44:36 +0200377 } else
378 lock_page(page);
Jens Axboe7480a902006-04-11 13:52:47 +0200379
380 /*
Miklos Szeredi32502b82008-07-04 09:35:17 +0200381 * Page was truncated, or invalidated by the
382 * filesystem. Redo the find/create, but this time the
383 * page is kept locked, so there's no chance of another
384 * race with truncate/invalidate.
Jens Axboe7480a902006-04-11 13:52:47 +0200385 */
386 if (!page->mapping) {
387 unlock_page(page);
Miklos Szeredi32502b82008-07-04 09:35:17 +0200388 page = find_or_create_page(mapping, index,
389 mapping_gfp_mask(mapping));
390
391 if (!page) {
392 error = -ENOMEM;
393 break;
394 }
395 page_cache_release(pages[page_nr]);
396 pages[page_nr] = page;
Jens Axboe7480a902006-04-11 13:52:47 +0200397 }
398 /*
399 * page was already under io and is now done, great
400 */
401 if (PageUptodate(page)) {
402 unlock_page(page);
403 goto fill_it;
404 }
405
Jens Axboe7480a902006-04-11 13:52:47 +0200406 /*
407 * need to read in the page
408 */
409 error = mapping->a_ops->readpage(in, page);
Jens Axboe7480a902006-04-11 13:52:47 +0200410 if (unlikely(error)) {
Jens Axboeeb207962006-04-27 11:05:22 +0200411 /*
412 * We really should re-lookup the page here,
413 * but it complicates things a lot. Instead
414 * lets just do what we already stored, and
415 * we'll get it the next time we are called.
416 */
Jens Axboe7480a902006-04-11 13:52:47 +0200417 if (error == AOP_TRUNCATED_PAGE)
Jens Axboeeb207962006-04-27 11:05:22 +0200418 error = 0;
419
Jens Axboe7480a902006-04-11 13:52:47 +0200420 break;
421 }
Jens Axboe620a3242007-06-07 09:39:42 +0200422 }
423fill_it:
424 /*
425 * i_size must be checked after PageUptodate.
426 */
427 isize = i_size_read(mapping->host);
428 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
429 if (unlikely(!isize || index > end_index))
430 break;
431
432 /*
433 * if this is the last page, see if we need to shrink
434 * the length and stop
435 */
436 if (end_index == index) {
437 unsigned int plen;
Jens Axboe91ad66e2006-04-19 15:55:10 +0200438
439 /*
Jens Axboe620a3242007-06-07 09:39:42 +0200440 * max good bytes in this page
Jens Axboe91ad66e2006-04-19 15:55:10 +0200441 */
Jens Axboe620a3242007-06-07 09:39:42 +0200442 plen = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
443 if (plen <= loff)
Jens Axboe91ad66e2006-04-19 15:55:10 +0200444 break;
Jens Axboe91ad66e2006-04-19 15:55:10 +0200445
446 /*
Jens Axboe620a3242007-06-07 09:39:42 +0200447 * force quit after adding this page
Jens Axboe91ad66e2006-04-19 15:55:10 +0200448 */
Jens Axboe620a3242007-06-07 09:39:42 +0200449 this_len = min(this_len, plen - loff);
450 len = this_len;
Jens Axboe7480a902006-04-11 13:52:47 +0200451 }
Jens Axboe620a3242007-06-07 09:39:42 +0200452
Jens Axboeeb207962006-04-27 11:05:22 +0200453 partial[page_nr].offset = loff;
454 partial[page_nr].len = this_len;
Jens Axboe82aa5d62006-04-20 13:05:48 +0200455 len -= this_len;
Jens Axboe91ad66e2006-04-19 15:55:10 +0200456 loff = 0;
Jens Axboeeb207962006-04-27 11:05:22 +0200457 spd.nr_pages++;
458 index++;
Jens Axboe5274f052006-03-30 15:15:30 +0200459 }
460
Jens Axboeeb207962006-04-27 11:05:22 +0200461 /*
Hugh Dickins475ecad2007-06-07 09:36:00 +0200462 * Release any pages at the end, if we quit early. 'page_nr' is how far
Jens Axboeeb207962006-04-27 11:05:22 +0200463 * we got, 'nr_pages' is how many pages are in the map.
464 */
465 while (page_nr < nr_pages)
466 page_cache_release(pages[page_nr++]);
Fengguang Wuf4e6b492007-10-16 01:24:33 -0700467 in->f_ra.prev_pos = (loff_t)index << PAGE_CACHE_SHIFT;
Jens Axboeeb207962006-04-27 11:05:22 +0200468
Jens Axboe912d35f2006-04-26 10:59:21 +0200469 if (spd.nr_pages)
Jens Axboe00522fb2006-04-26 14:39:29 +0200470 return splice_to_pipe(pipe, &spd);
Jens Axboe5274f052006-03-30 15:15:30 +0200471
Jens Axboe7480a902006-04-11 13:52:47 +0200472 return error;
Jens Axboe5274f052006-03-30 15:15:30 +0200473}
474
Jens Axboe83f91352006-04-02 23:05:09 +0200475/**
476 * generic_file_splice_read - splice data from file to a pipe
477 * @in: file to splice from
Jens Axboe932cc6d2007-06-21 13:10:21 +0200478 * @ppos: position in @in
Jens Axboe83f91352006-04-02 23:05:09 +0200479 * @pipe: pipe to splice to
480 * @len: number of bytes to splice
481 * @flags: splice modifier flags
482 *
Jens Axboe932cc6d2007-06-21 13:10:21 +0200483 * Description:
484 * Will read pages from given file and fill them into a pipe. Can be
485 * used as long as the address_space operations for the source implements
486 * a readpage() hook.
487 *
Jens Axboe83f91352006-04-02 23:05:09 +0200488 */
Jens Axboecbb7e572006-04-11 14:57:50 +0200489ssize_t generic_file_splice_read(struct file *in, loff_t *ppos,
490 struct pipe_inode_info *pipe, size_t len,
491 unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +0200492{
Jens Axboed366d3982007-06-01 14:54:11 +0200493 loff_t isize, left;
Jens Axboe8191ecd2008-04-10 08:24:25 +0200494 int ret;
Jens Axboed366d3982007-06-01 14:54:11 +0200495
496 isize = i_size_read(in->f_mapping->host);
497 if (unlikely(*ppos >= isize))
498 return 0;
499
500 left = isize - *ppos;
501 if (unlikely(left < len))
502 len = left;
Jens Axboe5274f052006-03-30 15:15:30 +0200503
Jens Axboe8191ecd2008-04-10 08:24:25 +0200504 ret = __generic_file_splice_read(in, ppos, pipe, len, flags);
505 if (ret > 0)
Jens Axboecbb7e572006-04-11 14:57:50 +0200506 *ppos += ret;
Jens Axboe5274f052006-03-30 15:15:30 +0200507
508 return ret;
509}
Jens Axboe059a8f32006-04-02 23:06:05 +0200510EXPORT_SYMBOL(generic_file_splice_read);
511
Miklos Szeredi68181732009-05-07 15:37:36 +0200512static const struct pipe_buf_operations default_pipe_buf_ops = {
513 .can_merge = 0,
514 .map = generic_pipe_buf_map,
515 .unmap = generic_pipe_buf_unmap,
516 .confirm = generic_pipe_buf_confirm,
517 .release = generic_pipe_buf_release,
518 .steal = generic_pipe_buf_steal,
519 .get = generic_pipe_buf_get,
520};
521
522static ssize_t kernel_readv(struct file *file, const struct iovec *vec,
523 unsigned long vlen, loff_t offset)
524{
525 mm_segment_t old_fs;
526 loff_t pos = offset;
527 ssize_t res;
528
529 old_fs = get_fs();
530 set_fs(get_ds());
531 /* The cast to a user pointer is valid due to the set_fs() */
532 res = vfs_readv(file, (const struct iovec __user *)vec, vlen, &pos);
533 set_fs(old_fs);
534
535 return res;
536}
537
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +0200538static ssize_t kernel_writev(struct file *file, const struct iovec *vec,
539 unsigned long vlen, loff_t *ppos)
540{
541 mm_segment_t old_fs;
542 ssize_t res;
543
544 old_fs = get_fs();
545 set_fs(get_ds());
546 /* The cast to a user pointer is valid due to the set_fs() */
547 res = vfs_writev(file, (const struct iovec __user *)vec, vlen, ppos);
548 set_fs(old_fs);
549
550 return res;
551}
552
Miklos Szeredi68181732009-05-07 15:37:36 +0200553ssize_t default_file_splice_read(struct file *in, loff_t *ppos,
554 struct pipe_inode_info *pipe, size_t len,
555 unsigned int flags)
556{
557 unsigned int nr_pages;
558 unsigned int nr_freed;
559 size_t offset;
560 struct page *pages[PIPE_BUFFERS];
561 struct partial_page partial[PIPE_BUFFERS];
562 struct iovec vec[PIPE_BUFFERS];
563 pgoff_t index;
564 ssize_t res;
565 size_t this_len;
566 int error;
567 int i;
568 struct splice_pipe_desc spd = {
569 .pages = pages,
570 .partial = partial,
571 .flags = flags,
572 .ops = &default_pipe_buf_ops,
573 .spd_release = spd_release_page,
574 };
575
576 index = *ppos >> PAGE_CACHE_SHIFT;
577 offset = *ppos & ~PAGE_CACHE_MASK;
578 nr_pages = (len + offset + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
579
580 for (i = 0; i < nr_pages && i < PIPE_BUFFERS && len; i++) {
581 struct page *page;
582
583 page = alloc_page(GFP_HIGHUSER);
584 error = -ENOMEM;
585 if (!page)
586 goto err;
587
588 this_len = min_t(size_t, len, PAGE_CACHE_SIZE - offset);
589 vec[i].iov_base = (void __user *) kmap(page);
590 vec[i].iov_len = this_len;
591 pages[i] = page;
592 spd.nr_pages++;
593 len -= this_len;
594 offset = 0;
595 }
596
597 res = kernel_readv(in, vec, spd.nr_pages, *ppos);
598 if (res < 0)
599 goto err;
600
601 error = 0;
602 if (!res)
603 goto err;
604
605 nr_freed = 0;
606 for (i = 0; i < spd.nr_pages; i++) {
607 kunmap(pages[i]);
608 this_len = min_t(size_t, vec[i].iov_len, res);
609 partial[i].offset = 0;
610 partial[i].len = this_len;
611 if (!this_len) {
612 __free_page(pages[i]);
613 pages[i] = NULL;
614 nr_freed++;
615 }
616 res -= this_len;
617 }
618 spd.nr_pages -= nr_freed;
619
620 res = splice_to_pipe(pipe, &spd);
621 if (res > 0)
622 *ppos += res;
623
624 return res;
625
626err:
627 for (i = 0; i < spd.nr_pages; i++) {
628 kunmap(pages[i]);
629 __free_page(pages[i]);
630 }
631 return error;
632}
633EXPORT_SYMBOL(default_file_splice_read);
634
Jens Axboe5274f052006-03-30 15:15:30 +0200635/*
Jens Axboe4f6f0bd2006-04-02 23:04:46 +0200636 * Send 'sd->len' bytes to socket from 'sd->file' at position 'sd->pos'
Jens Axboe016b6612006-04-25 15:42:00 +0200637 * using sendpage(). Return the number of bytes sent.
Jens Axboe5274f052006-03-30 15:15:30 +0200638 */
Jens Axboe76ad4d12006-05-03 10:41:33 +0200639static int pipe_to_sendpage(struct pipe_inode_info *pipe,
Jens Axboe5274f052006-03-30 15:15:30 +0200640 struct pipe_buffer *buf, struct splice_desc *sd)
641{
Jens Axboe6a14b902007-06-14 13:08:55 +0200642 struct file *file = sd->u.file;
Jens Axboe5274f052006-03-30 15:15:30 +0200643 loff_t pos = sd->pos;
Jens Axboef84d7512006-05-01 19:59:03 +0200644 int ret, more;
Jens Axboe5274f052006-03-30 15:15:30 +0200645
Jens Axboecac36bb2007-06-14 13:10:48 +0200646 ret = buf->ops->confirm(pipe, buf);
Jens Axboef84d7512006-05-01 19:59:03 +0200647 if (!ret) {
648 more = (sd->flags & SPLICE_F_MORE) || sd->len < sd->total_len;
Jens Axboe5274f052006-03-30 15:15:30 +0200649
Jens Axboef84d7512006-05-01 19:59:03 +0200650 ret = file->f_op->sendpage(file, buf->page, buf->offset,
651 sd->len, &pos, more);
652 }
Jens Axboe5274f052006-03-30 15:15:30 +0200653
Jens Axboe016b6612006-04-25 15:42:00 +0200654 return ret;
Jens Axboe5274f052006-03-30 15:15:30 +0200655}
656
657/*
658 * This is a little more tricky than the file -> pipe splicing. There are
659 * basically three cases:
660 *
661 * - Destination page already exists in the address space and there
662 * are users of it. For that case we have no other option that
663 * copying the data. Tough luck.
664 * - Destination page already exists in the address space, but there
665 * are no users of it. Make sure it's uptodate, then drop it. Fall
666 * through to last case.
667 * - Destination page does not exist, we can add the pipe page to
668 * the page cache and avoid the copy.
669 *
Jens Axboe83f91352006-04-02 23:05:09 +0200670 * If asked to move pages to the output file (SPLICE_F_MOVE is set in
671 * sd->flags), we attempt to migrate pages from the pipe to the output
672 * file address space page cache. This is possible if no one else has
673 * the pipe page referenced outside of the pipe and page cache. If
674 * SPLICE_F_MOVE isn't set, or we cannot move the page, we simply create
675 * a new page in the output file page cache and fill/dirty that.
Jens Axboe5274f052006-03-30 15:15:30 +0200676 */
Miklos Szeredi328eaab2009-04-14 19:48:39 +0200677int pipe_to_file(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
678 struct splice_desc *sd)
Jens Axboe5274f052006-03-30 15:15:30 +0200679{
Jens Axboe6a14b902007-06-14 13:08:55 +0200680 struct file *file = sd->u.file;
Jens Axboe5274f052006-03-30 15:15:30 +0200681 struct address_space *mapping = file->f_mapping;
Jens Axboe016b6612006-04-25 15:42:00 +0200682 unsigned int offset, this_len;
Jens Axboe5274f052006-03-30 15:15:30 +0200683 struct page *page;
Nick Pigginafddba42007-10-16 01:25:01 -0700684 void *fsdata;
Jens Axboe3e7ee3e2006-04-02 23:11:04 +0200685 int ret;
Jens Axboe5274f052006-03-30 15:15:30 +0200686
687 /*
Jens Axboe49d0b212006-04-10 09:04:41 +0200688 * make sure the data in this buffer is uptodate
Jens Axboe5274f052006-03-30 15:15:30 +0200689 */
Jens Axboecac36bb2007-06-14 13:10:48 +0200690 ret = buf->ops->confirm(pipe, buf);
Jens Axboef84d7512006-05-01 19:59:03 +0200691 if (unlikely(ret))
692 return ret;
Jens Axboe5274f052006-03-30 15:15:30 +0200693
Jens Axboe5274f052006-03-30 15:15:30 +0200694 offset = sd->pos & ~PAGE_CACHE_MASK;
695
Jens Axboe016b6612006-04-25 15:42:00 +0200696 this_len = sd->len;
697 if (this_len + offset > PAGE_CACHE_SIZE)
698 this_len = PAGE_CACHE_SIZE - offset;
699
Nick Pigginafddba42007-10-16 01:25:01 -0700700 ret = pagecache_write_begin(file, mapping, sd->pos, this_len,
701 AOP_FLAG_UNINTERRUPTIBLE, &page, &fsdata);
702 if (unlikely(ret))
703 goto out;
Jens Axboe5274f052006-03-30 15:15:30 +0200704
Jens Axboe0568b402006-05-01 19:50:48 +0200705 if (buf->page != page) {
Jens Axboef84d7512006-05-01 19:59:03 +0200706 /*
707 * Careful, ->map() uses KM_USER0!
708 */
Jens Axboe76ad4d12006-05-03 10:41:33 +0200709 char *src = buf->ops->map(pipe, buf, 1);
Jens Axboef84d7512006-05-01 19:59:03 +0200710 char *dst = kmap_atomic(page, KM_USER1);
Jens Axboe5abc97a2006-03-30 15:16:46 +0200711
Jens Axboe016b6612006-04-25 15:42:00 +0200712 memcpy(dst + offset, src + buf->offset, this_len);
Jens Axboe5abc97a2006-03-30 15:16:46 +0200713 flush_dcache_page(page);
Jens Axboef84d7512006-05-01 19:59:03 +0200714 kunmap_atomic(dst, KM_USER1);
Jens Axboe76ad4d12006-05-03 10:41:33 +0200715 buf->ops->unmap(pipe, buf, src);
Jens Axboe5abc97a2006-03-30 15:16:46 +0200716 }
Nick Pigginafddba42007-10-16 01:25:01 -0700717 ret = pagecache_write_end(file, mapping, sd->pos, this_len, this_len,
718 page, fsdata);
Jens Axboe5274f052006-03-30 15:15:30 +0200719out:
Jens Axboe5274f052006-03-30 15:15:30 +0200720 return ret;
721}
Miklos Szeredi328eaab2009-04-14 19:48:39 +0200722EXPORT_SYMBOL(pipe_to_file);
Jens Axboe5274f052006-03-30 15:15:30 +0200723
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200724static void wakeup_pipe_writers(struct pipe_inode_info *pipe)
725{
726 smp_mb();
727 if (waitqueue_active(&pipe->wait))
728 wake_up_interruptible(&pipe->wait);
729 kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
730}
731
732/**
733 * splice_from_pipe_feed - feed available data from a pipe to a file
734 * @pipe: pipe to splice from
735 * @sd: information to @actor
736 * @actor: handler that splices the data
737 *
738 * Description:
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200739 * This function loops over the pipe and calls @actor to do the
740 * actual moving of a single struct pipe_buffer to the desired
741 * destination. It returns when there's no more buffers left in
742 * the pipe or if the requested number of bytes (@sd->total_len)
743 * have been copied. It returns a positive number (one) if the
744 * pipe needs to be filled with more data, zero if the required
745 * number of bytes have been copied and -errno on error.
746 *
747 * This, together with splice_from_pipe_{begin,end,next}, may be
748 * used to implement the functionality of __splice_from_pipe() when
749 * locking is required around copying the pipe buffers to the
750 * destination.
751 */
752int splice_from_pipe_feed(struct pipe_inode_info *pipe, struct splice_desc *sd,
753 splice_actor *actor)
754{
755 int ret;
756
757 while (pipe->nrbufs) {
758 struct pipe_buffer *buf = pipe->bufs + pipe->curbuf;
759 const struct pipe_buf_operations *ops = buf->ops;
760
761 sd->len = buf->len;
762 if (sd->len > sd->total_len)
763 sd->len = sd->total_len;
764
765 ret = actor(pipe, buf, sd);
766 if (ret <= 0) {
767 if (ret == -ENODATA)
768 ret = 0;
769 return ret;
770 }
771 buf->offset += ret;
772 buf->len -= ret;
773
774 sd->num_spliced += ret;
775 sd->len -= ret;
776 sd->pos += ret;
777 sd->total_len -= ret;
778
779 if (!buf->len) {
780 buf->ops = NULL;
781 ops->release(pipe, buf);
782 pipe->curbuf = (pipe->curbuf + 1) & (PIPE_BUFFERS - 1);
783 pipe->nrbufs--;
784 if (pipe->inode)
785 sd->need_wakeup = true;
786 }
787
788 if (!sd->total_len)
789 return 0;
790 }
791
792 return 1;
793}
794EXPORT_SYMBOL(splice_from_pipe_feed);
795
796/**
797 * splice_from_pipe_next - wait for some data to splice from
798 * @pipe: pipe to splice from
799 * @sd: information about the splice operation
800 *
801 * Description:
802 * This function will wait for some data and return a positive
803 * value (one) if pipe buffers are available. It will return zero
804 * or -errno if no more data needs to be spliced.
805 */
806int splice_from_pipe_next(struct pipe_inode_info *pipe, struct splice_desc *sd)
807{
808 while (!pipe->nrbufs) {
809 if (!pipe->writers)
810 return 0;
811
812 if (!pipe->waiting_writers && sd->num_spliced)
813 return 0;
814
815 if (sd->flags & SPLICE_F_NONBLOCK)
816 return -EAGAIN;
817
818 if (signal_pending(current))
819 return -ERESTARTSYS;
820
821 if (sd->need_wakeup) {
822 wakeup_pipe_writers(pipe);
823 sd->need_wakeup = false;
824 }
825
826 pipe_wait(pipe);
827 }
828
829 return 1;
830}
831EXPORT_SYMBOL(splice_from_pipe_next);
832
833/**
834 * splice_from_pipe_begin - start splicing from pipe
Randy Dunlapb80901b2009-04-16 19:09:55 -0700835 * @sd: information about the splice operation
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200836 *
837 * Description:
838 * This function should be called before a loop containing
839 * splice_from_pipe_next() and splice_from_pipe_feed() to
840 * initialize the necessary fields of @sd.
841 */
842void splice_from_pipe_begin(struct splice_desc *sd)
843{
844 sd->num_spliced = 0;
845 sd->need_wakeup = false;
846}
847EXPORT_SYMBOL(splice_from_pipe_begin);
848
849/**
850 * splice_from_pipe_end - finish splicing from pipe
851 * @pipe: pipe to splice from
852 * @sd: information about the splice operation
853 *
854 * Description:
855 * This function will wake up pipe writers if necessary. It should
856 * be called after a loop containing splice_from_pipe_next() and
857 * splice_from_pipe_feed().
858 */
859void splice_from_pipe_end(struct pipe_inode_info *pipe, struct splice_desc *sd)
860{
861 if (sd->need_wakeup)
862 wakeup_pipe_writers(pipe);
863}
864EXPORT_SYMBOL(splice_from_pipe_end);
865
Jens Axboe932cc6d2007-06-21 13:10:21 +0200866/**
867 * __splice_from_pipe - splice data from a pipe to given actor
868 * @pipe: pipe to splice from
869 * @sd: information to @actor
870 * @actor: handler that splices the data
871 *
872 * Description:
873 * This function does little more than loop over the pipe and call
874 * @actor to do the actual moving of a single struct pipe_buffer to
875 * the desired destination. See pipe_to_file, pipe_to_sendpage, or
876 * pipe_to_user.
877 *
Jens Axboe83f91352006-04-02 23:05:09 +0200878 */
Jens Axboec66ab6f2007-06-12 21:17:17 +0200879ssize_t __splice_from_pipe(struct pipe_inode_info *pipe, struct splice_desc *sd,
880 splice_actor *actor)
Jens Axboe5274f052006-03-30 15:15:30 +0200881{
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200882 int ret;
Jens Axboe5274f052006-03-30 15:15:30 +0200883
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200884 splice_from_pipe_begin(sd);
885 do {
886 ret = splice_from_pipe_next(pipe, sd);
887 if (ret > 0)
888 ret = splice_from_pipe_feed(pipe, sd, actor);
889 } while (ret > 0);
890 splice_from_pipe_end(pipe, sd);
Jens Axboe5274f052006-03-30 15:15:30 +0200891
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200892 return sd->num_spliced ? sd->num_spliced : ret;
Jens Axboe5274f052006-03-30 15:15:30 +0200893}
Mark Fasheh40bee442007-03-21 13:11:02 +0100894EXPORT_SYMBOL(__splice_from_pipe);
Jens Axboe5274f052006-03-30 15:15:30 +0200895
Jens Axboe932cc6d2007-06-21 13:10:21 +0200896/**
897 * splice_from_pipe - splice data from a pipe to a file
898 * @pipe: pipe to splice from
899 * @out: file to splice to
900 * @ppos: position in @out
901 * @len: how many bytes to splice
902 * @flags: splice modifier flags
903 * @actor: handler that splices the data
904 *
905 * Description:
Miklos Szeredi29339702009-04-14 19:48:37 +0200906 * See __splice_from_pipe. This function locks the pipe inode,
Jens Axboe932cc6d2007-06-21 13:10:21 +0200907 * otherwise it's identical to __splice_from_pipe().
908 *
909 */
Mark Fasheh6da61802006-10-17 18:43:07 +0200910ssize_t splice_from_pipe(struct pipe_inode_info *pipe, struct file *out,
911 loff_t *ppos, size_t len, unsigned int flags,
912 splice_actor *actor)
913{
914 ssize_t ret;
Jens Axboec66ab6f2007-06-12 21:17:17 +0200915 struct splice_desc sd = {
916 .total_len = len,
917 .flags = flags,
918 .pos = *ppos,
Jens Axboe6a14b902007-06-14 13:08:55 +0200919 .u.file = out,
Jens Axboec66ab6f2007-06-12 21:17:17 +0200920 };
Mark Fasheh6da61802006-10-17 18:43:07 +0200921
Miklos Szeredi61e0d472009-04-14 19:48:41 +0200922 pipe_lock(pipe);
Jens Axboec66ab6f2007-06-12 21:17:17 +0200923 ret = __splice_from_pipe(pipe, &sd, actor);
Miklos Szeredi61e0d472009-04-14 19:48:41 +0200924 pipe_unlock(pipe);
Mark Fasheh6da61802006-10-17 18:43:07 +0200925
926 return ret;
927}
928
929/**
Jens Axboe83f91352006-04-02 23:05:09 +0200930 * generic_file_splice_write - splice data from a pipe to a file
Ingo Molnar3a326a22006-04-10 15:18:35 +0200931 * @pipe: pipe info
Jens Axboe83f91352006-04-02 23:05:09 +0200932 * @out: file to write to
Jens Axboe932cc6d2007-06-21 13:10:21 +0200933 * @ppos: position in @out
Jens Axboe83f91352006-04-02 23:05:09 +0200934 * @len: number of bytes to splice
935 * @flags: splice modifier flags
936 *
Jens Axboe932cc6d2007-06-21 13:10:21 +0200937 * Description:
938 * Will either move or copy pages (determined by @flags options) from
939 * the given pipe inode to the given file.
Jens Axboe83f91352006-04-02 23:05:09 +0200940 *
941 */
Ingo Molnar3a326a22006-04-10 15:18:35 +0200942ssize_t
943generic_file_splice_write(struct pipe_inode_info *pipe, struct file *out,
Jens Axboecbb7e572006-04-11 14:57:50 +0200944 loff_t *ppos, size_t len, unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +0200945{
Jens Axboe4f6f0bd2006-04-02 23:04:46 +0200946 struct address_space *mapping = out->f_mapping;
Jens Axboe8c34e2d2006-10-17 19:43:22 +0200947 struct inode *inode = mapping->host;
Miklos Szeredi7f3d4ee2008-05-07 09:22:39 +0200948 struct splice_desc sd = {
949 .total_len = len,
950 .flags = flags,
951 .pos = *ppos,
952 .u.file = out,
953 };
Ingo Molnar3a326a22006-04-10 15:18:35 +0200954 ssize_t ret;
Jens Axboe8c34e2d2006-10-17 19:43:22 +0200955
Miklos Szeredi61e0d472009-04-14 19:48:41 +0200956 pipe_lock(pipe);
Miklos Szeredieb443e52009-04-14 19:48:38 +0200957
958 splice_from_pipe_begin(&sd);
959 do {
960 ret = splice_from_pipe_next(pipe, &sd);
961 if (ret <= 0)
962 break;
963
964 mutex_lock_nested(&inode->i_mutex, I_MUTEX_CHILD);
965 ret = file_remove_suid(out);
966 if (!ret)
967 ret = splice_from_pipe_feed(pipe, &sd, pipe_to_file);
968 mutex_unlock(&inode->i_mutex);
969 } while (ret > 0);
970 splice_from_pipe_end(pipe, &sd);
971
Miklos Szeredi61e0d472009-04-14 19:48:41 +0200972 pipe_unlock(pipe);
Miklos Szeredieb443e52009-04-14 19:48:38 +0200973
974 if (sd.num_spliced)
975 ret = sd.num_spliced;
976
Jens Axboea4514eb2006-04-19 15:57:05 +0200977 if (ret > 0) {
Jens Axboe17ee4f42007-06-15 13:10:37 +0200978 unsigned long nr_pages;
979
Jens Axboea4514eb2006-04-19 15:57:05 +0200980 *ppos += ret;
Jens Axboe17ee4f42007-06-15 13:10:37 +0200981 nr_pages = (ret + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
Jens Axboe4f6f0bd2006-04-02 23:04:46 +0200982
Jens Axboea4514eb2006-04-19 15:57:05 +0200983 /*
984 * If file or inode is SYNC and we actually wrote some data,
985 * sync it.
986 */
987 if (unlikely((out->f_flags & O_SYNC) || IS_SYNC(inode))) {
Miklos Szeredi7f3d4ee2008-05-07 09:22:39 +0200988 int err;
989
Jens Axboea4514eb2006-04-19 15:57:05 +0200990 mutex_lock(&inode->i_mutex);
991 err = generic_osync_inode(inode, mapping,
992 OSYNC_METADATA|OSYNC_DATA);
993 mutex_unlock(&inode->i_mutex);
994
995 if (err)
996 ret = err;
997 }
Jens Axboe17ee4f42007-06-15 13:10:37 +0200998 balance_dirty_pages_ratelimited_nr(mapping, nr_pages);
Jens Axboe4f6f0bd2006-04-02 23:04:46 +0200999 }
1000
1001 return ret;
Jens Axboe5274f052006-03-30 15:15:30 +02001002}
1003
Jens Axboe059a8f32006-04-02 23:06:05 +02001004EXPORT_SYMBOL(generic_file_splice_write);
1005
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +02001006static struct pipe_buffer *nth_pipe_buf(struct pipe_inode_info *pipe, int n)
1007{
1008 return &pipe->bufs[(pipe->curbuf + n) % PIPE_BUFFERS];
1009}
1010
1011static ssize_t default_file_splice_write(struct pipe_inode_info *pipe,
1012 struct file *out, loff_t *ppos,
1013 size_t len, unsigned int flags)
1014{
1015 ssize_t ret = 0;
1016 ssize_t total_len = 0;
1017 int do_wakeup = 0;
1018
1019 pipe_lock(pipe);
1020 while (len) {
1021 struct pipe_buffer *buf;
1022 void *data[PIPE_BUFFERS];
1023 struct iovec vec[PIPE_BUFFERS];
1024 unsigned int nr_pages = 0;
1025 unsigned int write_len = 0;
1026 unsigned int now_len = len;
1027 unsigned int this_len;
1028 int i;
1029
1030 BUG_ON(pipe->nrbufs > PIPE_BUFFERS);
1031 for (i = 0; i < pipe->nrbufs && now_len; i++) {
1032 buf = nth_pipe_buf(pipe, i);
1033
1034 ret = buf->ops->confirm(pipe, buf);
1035 if (ret)
1036 break;
1037
1038 data[i] = buf->ops->map(pipe, buf, 0);
1039 this_len = min(buf->len, now_len);
1040 vec[i].iov_base = (void __user *) data[i] + buf->offset;
1041 vec[i].iov_len = this_len;
1042 now_len -= this_len;
1043 write_len += this_len;
1044 nr_pages++;
1045 }
1046
1047 if (nr_pages) {
1048 ret = kernel_writev(out, vec, nr_pages, ppos);
1049 if (ret == 0)
1050 ret = -EIO;
1051 if (ret > 0) {
1052 len -= ret;
1053 total_len += ret;
1054 }
1055 }
1056
1057 for (i = 0; i < nr_pages; i++) {
1058 buf = nth_pipe_buf(pipe, i);
1059 buf->ops->unmap(pipe, buf, data[i]);
1060
1061 if (ret > 0) {
1062 this_len = min_t(unsigned, vec[i].iov_len, ret);
1063 buf->offset += this_len;
1064 buf->len -= this_len;
1065 ret -= this_len;
1066 }
1067 }
1068
1069 if (ret < 0)
1070 break;
1071
1072 while (pipe->nrbufs) {
1073 const struct pipe_buf_operations *ops;
1074
1075 buf = nth_pipe_buf(pipe, 0);
1076 if (buf->len)
1077 break;
1078
1079 ops = buf->ops;
1080 buf->ops = NULL;
1081 ops->release(pipe, buf);
1082 pipe->curbuf = (pipe->curbuf + 1) % PIPE_BUFFERS;
1083 pipe->nrbufs--;
1084 if (pipe->inode)
1085 do_wakeup = 1;
1086 }
1087
1088 if (pipe->nrbufs)
1089 continue;
1090 if (!pipe->writers)
1091 break;
1092 if (!pipe->waiting_writers) {
1093 if (total_len)
1094 break;
1095 }
1096
1097 if (flags & SPLICE_F_NONBLOCK) {
1098 ret = -EAGAIN;
1099 break;
1100 }
1101
1102 if (signal_pending(current)) {
1103 ret = -ERESTARTSYS;
1104 break;
1105 }
1106
1107 if (do_wakeup) {
1108 wakeup_pipe_writers(pipe);
1109 do_wakeup = 0;
1110 }
1111
1112 pipe_wait(pipe);
1113 }
1114 pipe_unlock(pipe);
1115
1116 if (do_wakeup)
1117 wakeup_pipe_writers(pipe);
1118
1119 return total_len ? total_len : ret;
1120}
1121
Jens Axboe83f91352006-04-02 23:05:09 +02001122/**
1123 * generic_splice_sendpage - splice data from a pipe to a socket
Jens Axboe932cc6d2007-06-21 13:10:21 +02001124 * @pipe: pipe to splice from
Jens Axboe83f91352006-04-02 23:05:09 +02001125 * @out: socket to write to
Jens Axboe932cc6d2007-06-21 13:10:21 +02001126 * @ppos: position in @out
Jens Axboe83f91352006-04-02 23:05:09 +02001127 * @len: number of bytes to splice
1128 * @flags: splice modifier flags
1129 *
Jens Axboe932cc6d2007-06-21 13:10:21 +02001130 * Description:
1131 * Will send @len bytes from the pipe to a network socket. No data copying
1132 * is involved.
Jens Axboe83f91352006-04-02 23:05:09 +02001133 *
1134 */
Ingo Molnar3a326a22006-04-10 15:18:35 +02001135ssize_t generic_splice_sendpage(struct pipe_inode_info *pipe, struct file *out,
Jens Axboecbb7e572006-04-11 14:57:50 +02001136 loff_t *ppos, size_t len, unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +02001137{
Jens Axboe00522fb2006-04-26 14:39:29 +02001138 return splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_sendpage);
Jens Axboe5274f052006-03-30 15:15:30 +02001139}
1140
Jens Axboe059a8f32006-04-02 23:06:05 +02001141EXPORT_SYMBOL(generic_splice_sendpage);
Jeff Garzika0f06782006-03-30 23:06:13 -05001142
Jens Axboe83f91352006-04-02 23:05:09 +02001143/*
1144 * Attempt to initiate a splice from pipe to file.
1145 */
Ingo Molnar3a326a22006-04-10 15:18:35 +02001146static long do_splice_from(struct pipe_inode_info *pipe, struct file *out,
Jens Axboecbb7e572006-04-11 14:57:50 +02001147 loff_t *ppos, size_t len, unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +02001148{
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +02001149 ssize_t (*splice_write)(struct pipe_inode_info *, struct file *,
1150 loff_t *, size_t, unsigned int);
Jens Axboe5274f052006-03-30 15:15:30 +02001151 int ret;
1152
Jens Axboe49570e92006-04-11 13:56:09 +02001153 if (unlikely(!(out->f_mode & FMODE_WRITE)))
Jens Axboe5274f052006-03-30 15:15:30 +02001154 return -EBADF;
1155
Linus Torvaldsefc968d2008-10-09 14:04:54 -07001156 if (unlikely(out->f_flags & O_APPEND))
1157 return -EINVAL;
1158
Jens Axboecbb7e572006-04-11 14:57:50 +02001159 ret = rw_verify_area(WRITE, out, ppos, len);
Jens Axboe5274f052006-03-30 15:15:30 +02001160 if (unlikely(ret < 0))
1161 return ret;
1162
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +02001163 splice_write = out->f_op->splice_write;
1164 if (!splice_write)
1165 splice_write = default_file_splice_write;
1166
1167 return splice_write(pipe, out, ppos, len, flags);
Jens Axboe5274f052006-03-30 15:15:30 +02001168}
1169
Jens Axboe83f91352006-04-02 23:05:09 +02001170/*
1171 * Attempt to initiate a splice from a file to a pipe.
1172 */
Jens Axboecbb7e572006-04-11 14:57:50 +02001173static long do_splice_to(struct file *in, loff_t *ppos,
1174 struct pipe_inode_info *pipe, size_t len,
1175 unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +02001176{
Miklos Szeredi68181732009-05-07 15:37:36 +02001177 ssize_t (*splice_read)(struct file *, loff_t *,
1178 struct pipe_inode_info *, size_t, unsigned int);
Jens Axboe5274f052006-03-30 15:15:30 +02001179 int ret;
1180
Jens Axboe49570e92006-04-11 13:56:09 +02001181 if (unlikely(!(in->f_mode & FMODE_READ)))
Jens Axboe5274f052006-03-30 15:15:30 +02001182 return -EBADF;
1183
Jens Axboecbb7e572006-04-11 14:57:50 +02001184 ret = rw_verify_area(READ, in, ppos, len);
Jens Axboe5274f052006-03-30 15:15:30 +02001185 if (unlikely(ret < 0))
1186 return ret;
1187
Miklos Szeredi68181732009-05-07 15:37:36 +02001188 splice_read = in->f_op->splice_read;
1189 if (!splice_read)
1190 splice_read = default_file_splice_read;
1191
1192 return splice_read(in, ppos, pipe, len, flags);
Jens Axboe5274f052006-03-30 15:15:30 +02001193}
1194
Jens Axboe932cc6d2007-06-21 13:10:21 +02001195/**
1196 * splice_direct_to_actor - splices data directly between two non-pipes
1197 * @in: file to splice from
1198 * @sd: actor information on where to splice to
1199 * @actor: handles the data splicing
1200 *
1201 * Description:
1202 * This is a special case helper to splice directly between two
1203 * points, without requiring an explicit pipe. Internally an allocated
Randy Dunlap79685b82007-07-27 08:08:51 +02001204 * pipe is cached in the process, and reused during the lifetime of
Jens Axboe932cc6d2007-06-21 13:10:21 +02001205 * that process.
1206 *
Jens Axboec66ab6f2007-06-12 21:17:17 +02001207 */
1208ssize_t splice_direct_to_actor(struct file *in, struct splice_desc *sd,
1209 splice_direct_actor *actor)
Jens Axboeb92ce552006-04-11 13:52:07 +02001210{
1211 struct pipe_inode_info *pipe;
1212 long ret, bytes;
1213 umode_t i_mode;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001214 size_t len;
1215 int i, flags;
Jens Axboeb92ce552006-04-11 13:52:07 +02001216
1217 /*
1218 * We require the input being a regular file, as we don't want to
1219 * randomly drop data for eg socket -> socket splicing. Use the
1220 * piped splicing for that!
1221 */
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001222 i_mode = in->f_path.dentry->d_inode->i_mode;
Jens Axboeb92ce552006-04-11 13:52:07 +02001223 if (unlikely(!S_ISREG(i_mode) && !S_ISBLK(i_mode)))
1224 return -EINVAL;
1225
1226 /*
1227 * neither in nor out is a pipe, setup an internal pipe attached to
1228 * 'out' and transfer the wanted data from 'in' to 'out' through that
1229 */
1230 pipe = current->splice_pipe;
Jens Axboe49570e92006-04-11 13:56:09 +02001231 if (unlikely(!pipe)) {
Jens Axboeb92ce552006-04-11 13:52:07 +02001232 pipe = alloc_pipe_info(NULL);
1233 if (!pipe)
1234 return -ENOMEM;
1235
1236 /*
1237 * We don't have an immediate reader, but we'll read the stuff
Jens Axboe00522fb2006-04-26 14:39:29 +02001238 * out of the pipe right after the splice_to_pipe(). So set
Jens Axboeb92ce552006-04-11 13:52:07 +02001239 * PIPE_READERS appropriately.
1240 */
1241 pipe->readers = 1;
1242
1243 current->splice_pipe = pipe;
1244 }
1245
1246 /*
Ingo Molnar73d62d82006-04-11 13:57:21 +02001247 * Do the splice.
Jens Axboeb92ce552006-04-11 13:52:07 +02001248 */
1249 ret = 0;
1250 bytes = 0;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001251 len = sd->total_len;
1252 flags = sd->flags;
1253
1254 /*
1255 * Don't block on output, we have to drain the direct pipe.
1256 */
1257 sd->flags &= ~SPLICE_F_NONBLOCK;
Jens Axboeb92ce552006-04-11 13:52:07 +02001258
1259 while (len) {
Jens Axboe51a92c02007-07-13 14:11:43 +02001260 size_t read_len;
Tom Zanussia82c53a2008-05-09 13:28:36 +02001261 loff_t pos = sd->pos, prev_pos = pos;
Jens Axboeb92ce552006-04-11 13:52:07 +02001262
Jens Axboebcd4f3a2007-07-16 14:41:49 +02001263 ret = do_splice_to(in, &pos, pipe, len, flags);
Jens Axboe51a92c02007-07-13 14:11:43 +02001264 if (unlikely(ret <= 0))
Jens Axboeb92ce552006-04-11 13:52:07 +02001265 goto out_release;
1266
1267 read_len = ret;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001268 sd->total_len = read_len;
Jens Axboeb92ce552006-04-11 13:52:07 +02001269
1270 /*
1271 * NOTE: nonblocking mode only applies to the input. We
1272 * must not do the output in nonblocking mode as then we
1273 * could get stuck data in the internal pipe:
1274 */
Jens Axboec66ab6f2007-06-12 21:17:17 +02001275 ret = actor(pipe, sd);
Tom Zanussia82c53a2008-05-09 13:28:36 +02001276 if (unlikely(ret <= 0)) {
1277 sd->pos = prev_pos;
Jens Axboeb92ce552006-04-11 13:52:07 +02001278 goto out_release;
Tom Zanussia82c53a2008-05-09 13:28:36 +02001279 }
Jens Axboeb92ce552006-04-11 13:52:07 +02001280
1281 bytes += ret;
1282 len -= ret;
Jens Axboebcd4f3a2007-07-16 14:41:49 +02001283 sd->pos = pos;
Jens Axboeb92ce552006-04-11 13:52:07 +02001284
Tom Zanussia82c53a2008-05-09 13:28:36 +02001285 if (ret < read_len) {
1286 sd->pos = prev_pos + ret;
Jens Axboe51a92c02007-07-13 14:11:43 +02001287 goto out_release;
Tom Zanussia82c53a2008-05-09 13:28:36 +02001288 }
Jens Axboeb92ce552006-04-11 13:52:07 +02001289 }
1290
Jens Axboe9e971982008-01-29 21:05:57 +01001291done:
Jens Axboeb92ce552006-04-11 13:52:07 +02001292 pipe->nrbufs = pipe->curbuf = 0;
Jens Axboe80848702008-01-30 12:24:48 +01001293 file_accessed(in);
Jens Axboeb92ce552006-04-11 13:52:07 +02001294 return bytes;
1295
1296out_release:
1297 /*
1298 * If we did an incomplete transfer we must release
1299 * the pipe buffers in question:
1300 */
1301 for (i = 0; i < PIPE_BUFFERS; i++) {
1302 struct pipe_buffer *buf = pipe->bufs + i;
1303
1304 if (buf->ops) {
1305 buf->ops->release(pipe, buf);
1306 buf->ops = NULL;
1307 }
1308 }
Jens Axboeb92ce552006-04-11 13:52:07 +02001309
Jens Axboe9e971982008-01-29 21:05:57 +01001310 if (!bytes)
1311 bytes = ret;
Jens Axboeb92ce552006-04-11 13:52:07 +02001312
Jens Axboe9e971982008-01-29 21:05:57 +01001313 goto done;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001314}
1315EXPORT_SYMBOL(splice_direct_to_actor);
1316
1317static int direct_splice_actor(struct pipe_inode_info *pipe,
1318 struct splice_desc *sd)
1319{
Jens Axboe6a14b902007-06-14 13:08:55 +02001320 struct file *file = sd->u.file;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001321
1322 return do_splice_from(pipe, file, &sd->pos, sd->total_len, sd->flags);
1323}
1324
Jens Axboe932cc6d2007-06-21 13:10:21 +02001325/**
1326 * do_splice_direct - splices data directly between two files
1327 * @in: file to splice from
1328 * @ppos: input file offset
1329 * @out: file to splice to
1330 * @len: number of bytes to splice
1331 * @flags: splice modifier flags
1332 *
1333 * Description:
1334 * For use by do_sendfile(). splice can easily emulate sendfile, but
1335 * doing it in the application would incur an extra system call
1336 * (splice in + splice out, as compared to just sendfile()). So this helper
1337 * can splice directly through a process-private pipe.
1338 *
1339 */
Jens Axboec66ab6f2007-06-12 21:17:17 +02001340long do_splice_direct(struct file *in, loff_t *ppos, struct file *out,
1341 size_t len, unsigned int flags)
1342{
1343 struct splice_desc sd = {
1344 .len = len,
1345 .total_len = len,
1346 .flags = flags,
1347 .pos = *ppos,
Jens Axboe6a14b902007-06-14 13:08:55 +02001348 .u.file = out,
Jens Axboec66ab6f2007-06-12 21:17:17 +02001349 };
Jens Axboe51a92c02007-07-13 14:11:43 +02001350 long ret;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001351
1352 ret = splice_direct_to_actor(in, &sd, direct_splice_actor);
Jens Axboe51a92c02007-07-13 14:11:43 +02001353 if (ret > 0)
Tom Zanussia82c53a2008-05-09 13:28:36 +02001354 *ppos = sd.pos;
Jens Axboe51a92c02007-07-13 14:11:43 +02001355
Jens Axboec66ab6f2007-06-12 21:17:17 +02001356 return ret;
Jens Axboeb92ce552006-04-11 13:52:07 +02001357}
1358
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001359static int splice_pipe_to_pipe(struct pipe_inode_info *ipipe,
1360 struct pipe_inode_info *opipe,
1361 size_t len, unsigned int flags);
Jens Axboe83f91352006-04-02 23:05:09 +02001362/*
Jens Axboeddac0d32006-11-04 12:49:32 +01001363 * After the inode slimming patch, i_pipe/i_bdev/i_cdev share the same
1364 * location, so checking ->i_pipe is not enough to verify that this is a
1365 * pipe.
1366 */
1367static inline struct pipe_inode_info *pipe_info(struct inode *inode)
1368{
1369 if (S_ISFIFO(inode->i_mode))
1370 return inode->i_pipe;
1371
1372 return NULL;
1373}
1374
1375/*
Jens Axboe83f91352006-04-02 23:05:09 +02001376 * Determine where to splice to/from.
1377 */
Ingo Molnar529565d2006-04-10 15:18:58 +02001378static long do_splice(struct file *in, loff_t __user *off_in,
1379 struct file *out, loff_t __user *off_out,
1380 size_t len, unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +02001381{
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001382 struct pipe_inode_info *ipipe;
1383 struct pipe_inode_info *opipe;
Jens Axboecbb7e572006-04-11 14:57:50 +02001384 loff_t offset, *off;
Jens Axboea4514eb2006-04-19 15:57:05 +02001385 long ret;
Jens Axboe5274f052006-03-30 15:15:30 +02001386
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001387 ipipe = pipe_info(in->f_path.dentry->d_inode);
1388 opipe = pipe_info(out->f_path.dentry->d_inode);
1389
1390 if (ipipe && opipe) {
1391 if (off_in || off_out)
1392 return -ESPIPE;
1393
1394 if (!(in->f_mode & FMODE_READ))
1395 return -EBADF;
1396
1397 if (!(out->f_mode & FMODE_WRITE))
1398 return -EBADF;
1399
1400 /* Splicing to self would be fun, but... */
1401 if (ipipe == opipe)
1402 return -EINVAL;
1403
1404 return splice_pipe_to_pipe(ipipe, opipe, len, flags);
1405 }
1406
1407 if (ipipe) {
Ingo Molnar529565d2006-04-10 15:18:58 +02001408 if (off_in)
1409 return -ESPIPE;
Jens Axboeb92ce552006-04-11 13:52:07 +02001410 if (off_out) {
1411 if (out->f_op->llseek == no_llseek)
1412 return -EINVAL;
Jens Axboecbb7e572006-04-11 14:57:50 +02001413 if (copy_from_user(&offset, off_out, sizeof(loff_t)))
Jens Axboeb92ce552006-04-11 13:52:07 +02001414 return -EFAULT;
Jens Axboecbb7e572006-04-11 14:57:50 +02001415 off = &offset;
1416 } else
1417 off = &out->f_pos;
Ingo Molnar529565d2006-04-10 15:18:58 +02001418
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001419 ret = do_splice_from(ipipe, out, off, len, flags);
Jens Axboea4514eb2006-04-19 15:57:05 +02001420
1421 if (off_out && copy_to_user(off_out, off, sizeof(loff_t)))
1422 ret = -EFAULT;
1423
1424 return ret;
Ingo Molnar529565d2006-04-10 15:18:58 +02001425 }
Jens Axboe5274f052006-03-30 15:15:30 +02001426
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001427 if (opipe) {
Ingo Molnar529565d2006-04-10 15:18:58 +02001428 if (off_out)
1429 return -ESPIPE;
Jens Axboeb92ce552006-04-11 13:52:07 +02001430 if (off_in) {
1431 if (in->f_op->llseek == no_llseek)
1432 return -EINVAL;
Jens Axboecbb7e572006-04-11 14:57:50 +02001433 if (copy_from_user(&offset, off_in, sizeof(loff_t)))
Jens Axboeb92ce552006-04-11 13:52:07 +02001434 return -EFAULT;
Jens Axboecbb7e572006-04-11 14:57:50 +02001435 off = &offset;
1436 } else
1437 off = &in->f_pos;
Ingo Molnar529565d2006-04-10 15:18:58 +02001438
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001439 ret = do_splice_to(in, off, opipe, len, flags);
Jens Axboea4514eb2006-04-19 15:57:05 +02001440
1441 if (off_in && copy_to_user(off_in, off, sizeof(loff_t)))
1442 ret = -EFAULT;
1443
1444 return ret;
Ingo Molnar529565d2006-04-10 15:18:58 +02001445 }
Jens Axboe5274f052006-03-30 15:15:30 +02001446
1447 return -EINVAL;
1448}
1449
Jens Axboe912d35f2006-04-26 10:59:21 +02001450/*
1451 * Map an iov into an array of pages and offset/length tupples. With the
1452 * partial_page structure, we can map several non-contiguous ranges into
1453 * our ones pages[] map instead of splitting that operation into pieces.
1454 * Could easily be exported as a generic helper for other users, in which
1455 * case one would probably want to add a 'max_nr_pages' parameter as well.
1456 */
1457static int get_iovec_page_array(const struct iovec __user *iov,
1458 unsigned int nr_vecs, struct page **pages,
Jens Axboe7afa6fd2006-05-01 20:02:33 +02001459 struct partial_page *partial, int aligned)
Jens Axboe912d35f2006-04-26 10:59:21 +02001460{
1461 int buffers = 0, error = 0;
1462
Jens Axboe912d35f2006-04-26 10:59:21 +02001463 while (nr_vecs) {
1464 unsigned long off, npages;
Linus Torvalds75723952007-10-01 13:17:28 -07001465 struct iovec entry;
Jens Axboe912d35f2006-04-26 10:59:21 +02001466 void __user *base;
1467 size_t len;
1468 int i;
1469
Linus Torvalds75723952007-10-01 13:17:28 -07001470 error = -EFAULT;
Nick Pigginbc40d732008-07-25 19:45:26 -07001471 if (copy_from_user(&entry, iov, sizeof(entry)))
Jens Axboe912d35f2006-04-26 10:59:21 +02001472 break;
Linus Torvalds75723952007-10-01 13:17:28 -07001473
1474 base = entry.iov_base;
1475 len = entry.iov_len;
Jens Axboe912d35f2006-04-26 10:59:21 +02001476
1477 /*
1478 * Sanity check this iovec. 0 read succeeds.
1479 */
Linus Torvalds75723952007-10-01 13:17:28 -07001480 error = 0;
Jens Axboe912d35f2006-04-26 10:59:21 +02001481 if (unlikely(!len))
1482 break;
1483 error = -EFAULT;
Bastian Blank712a30e2008-02-10 16:47:57 +02001484 if (!access_ok(VERIFY_READ, base, len))
Jens Axboe912d35f2006-04-26 10:59:21 +02001485 break;
1486
1487 /*
1488 * Get this base offset and number of pages, then map
1489 * in the user pages.
1490 */
1491 off = (unsigned long) base & ~PAGE_MASK;
Jens Axboe7afa6fd2006-05-01 20:02:33 +02001492
1493 /*
1494 * If asked for alignment, the offset must be zero and the
1495 * length a multiple of the PAGE_SIZE.
1496 */
1497 error = -EINVAL;
1498 if (aligned && (off || len & ~PAGE_MASK))
1499 break;
1500
Jens Axboe912d35f2006-04-26 10:59:21 +02001501 npages = (off + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1502 if (npages > PIPE_BUFFERS - buffers)
1503 npages = PIPE_BUFFERS - buffers;
1504
Nick Pigginbc40d732008-07-25 19:45:26 -07001505 error = get_user_pages_fast((unsigned long)base, npages,
1506 0, &pages[buffers]);
Jens Axboe912d35f2006-04-26 10:59:21 +02001507
1508 if (unlikely(error <= 0))
1509 break;
1510
1511 /*
1512 * Fill this contiguous range into the partial page map.
1513 */
1514 for (i = 0; i < error; i++) {
Jens Axboe75914892006-05-02 12:57:18 +02001515 const int plen = min_t(size_t, len, PAGE_SIZE - off);
Jens Axboe912d35f2006-04-26 10:59:21 +02001516
1517 partial[buffers].offset = off;
1518 partial[buffers].len = plen;
1519
1520 off = 0;
1521 len -= plen;
1522 buffers++;
1523 }
1524
1525 /*
1526 * We didn't complete this iov, stop here since it probably
1527 * means we have to move some of this into a pipe to
1528 * be able to continue.
1529 */
1530 if (len)
1531 break;
1532
1533 /*
1534 * Don't continue if we mapped fewer pages than we asked for,
1535 * or if we mapped the max number of pages that we have
1536 * room for.
1537 */
1538 if (error < npages || buffers == PIPE_BUFFERS)
1539 break;
1540
1541 nr_vecs--;
1542 iov++;
1543 }
1544
Jens Axboe912d35f2006-04-26 10:59:21 +02001545 if (buffers)
1546 return buffers;
1547
1548 return error;
1549}
1550
Jens Axboe6a14b902007-06-14 13:08:55 +02001551static int pipe_to_user(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
1552 struct splice_desc *sd)
1553{
1554 char *src;
1555 int ret;
1556
Jens Axboecac36bb2007-06-14 13:10:48 +02001557 ret = buf->ops->confirm(pipe, buf);
Jens Axboe6a14b902007-06-14 13:08:55 +02001558 if (unlikely(ret))
1559 return ret;
1560
1561 /*
1562 * See if we can use the atomic maps, by prefaulting in the
1563 * pages and doing an atomic copy
1564 */
1565 if (!fault_in_pages_writeable(sd->u.userptr, sd->len)) {
1566 src = buf->ops->map(pipe, buf, 1);
1567 ret = __copy_to_user_inatomic(sd->u.userptr, src + buf->offset,
1568 sd->len);
1569 buf->ops->unmap(pipe, buf, src);
1570 if (!ret) {
1571 ret = sd->len;
1572 goto out;
1573 }
1574 }
1575
1576 /*
1577 * No dice, use slow non-atomic map and copy
1578 */
1579 src = buf->ops->map(pipe, buf, 0);
1580
1581 ret = sd->len;
1582 if (copy_to_user(sd->u.userptr, src + buf->offset, sd->len))
1583 ret = -EFAULT;
1584
Jens Axboe6866bef2007-10-16 10:01:29 +02001585 buf->ops->unmap(pipe, buf, src);
Jens Axboe6a14b902007-06-14 13:08:55 +02001586out:
1587 if (ret > 0)
1588 sd->u.userptr += ret;
Jens Axboe6a14b902007-06-14 13:08:55 +02001589 return ret;
1590}
1591
1592/*
1593 * For lack of a better implementation, implement vmsplice() to userspace
1594 * as a simple copy of the pipes pages to the user iov.
1595 */
1596static long vmsplice_to_user(struct file *file, const struct iovec __user *iov,
1597 unsigned long nr_segs, unsigned int flags)
1598{
1599 struct pipe_inode_info *pipe;
1600 struct splice_desc sd;
1601 ssize_t size;
1602 int error;
1603 long ret;
1604
1605 pipe = pipe_info(file->f_path.dentry->d_inode);
1606 if (!pipe)
1607 return -EBADF;
1608
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001609 pipe_lock(pipe);
Jens Axboe6a14b902007-06-14 13:08:55 +02001610
1611 error = ret = 0;
1612 while (nr_segs) {
1613 void __user *base;
1614 size_t len;
1615
1616 /*
1617 * Get user address base and length for this iovec.
1618 */
1619 error = get_user(base, &iov->iov_base);
1620 if (unlikely(error))
1621 break;
1622 error = get_user(len, &iov->iov_len);
1623 if (unlikely(error))
1624 break;
1625
1626 /*
1627 * Sanity check this iovec. 0 read succeeds.
1628 */
1629 if (unlikely(!len))
1630 break;
1631 if (unlikely(!base)) {
1632 error = -EFAULT;
1633 break;
1634 }
1635
Jens Axboe88119302008-02-08 08:49:14 -08001636 if (unlikely(!access_ok(VERIFY_WRITE, base, len))) {
1637 error = -EFAULT;
1638 break;
1639 }
1640
Jens Axboe6a14b902007-06-14 13:08:55 +02001641 sd.len = 0;
1642 sd.total_len = len;
1643 sd.flags = flags;
1644 sd.u.userptr = base;
1645 sd.pos = 0;
1646
1647 size = __splice_from_pipe(pipe, &sd, pipe_to_user);
1648 if (size < 0) {
1649 if (!ret)
1650 ret = size;
1651
1652 break;
1653 }
1654
1655 ret += size;
1656
1657 if (size < len)
1658 break;
1659
1660 nr_segs--;
1661 iov++;
1662 }
1663
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001664 pipe_unlock(pipe);
Jens Axboe6a14b902007-06-14 13:08:55 +02001665
1666 if (!ret)
1667 ret = error;
1668
1669 return ret;
1670}
1671
Jens Axboe912d35f2006-04-26 10:59:21 +02001672/*
1673 * vmsplice splices a user address range into a pipe. It can be thought of
1674 * as splice-from-memory, where the regular splice is splice-from-file (or
1675 * to file). In both cases the output is a pipe, naturally.
Jens Axboe912d35f2006-04-26 10:59:21 +02001676 */
Jens Axboe6a14b902007-06-14 13:08:55 +02001677static long vmsplice_to_pipe(struct file *file, const struct iovec __user *iov,
1678 unsigned long nr_segs, unsigned int flags)
Jens Axboe912d35f2006-04-26 10:59:21 +02001679{
Jens Axboeddac0d32006-11-04 12:49:32 +01001680 struct pipe_inode_info *pipe;
Jens Axboe912d35f2006-04-26 10:59:21 +02001681 struct page *pages[PIPE_BUFFERS];
1682 struct partial_page partial[PIPE_BUFFERS];
1683 struct splice_pipe_desc spd = {
1684 .pages = pages,
1685 .partial = partial,
1686 .flags = flags,
1687 .ops = &user_page_pipe_buf_ops,
Jens Axboebbdfc2f2007-11-06 23:29:47 -08001688 .spd_release = spd_release_page,
Jens Axboe912d35f2006-04-26 10:59:21 +02001689 };
1690
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001691 pipe = pipe_info(file->f_path.dentry->d_inode);
Jens Axboeddac0d32006-11-04 12:49:32 +01001692 if (!pipe)
Jens Axboe912d35f2006-04-26 10:59:21 +02001693 return -EBADF;
Jens Axboe912d35f2006-04-26 10:59:21 +02001694
Jens Axboe7afa6fd2006-05-01 20:02:33 +02001695 spd.nr_pages = get_iovec_page_array(iov, nr_segs, pages, partial,
1696 flags & SPLICE_F_GIFT);
Jens Axboe912d35f2006-04-26 10:59:21 +02001697 if (spd.nr_pages <= 0)
1698 return spd.nr_pages;
1699
Jens Axboe00522fb2006-04-26 14:39:29 +02001700 return splice_to_pipe(pipe, &spd);
Jens Axboe912d35f2006-04-26 10:59:21 +02001701}
1702
Jens Axboe6a14b902007-06-14 13:08:55 +02001703/*
1704 * Note that vmsplice only really supports true splicing _from_ user memory
1705 * to a pipe, not the other way around. Splicing from user memory is a simple
1706 * operation that can be supported without any funky alignment restrictions
1707 * or nasty vm tricks. We simply map in the user memory and fill them into
1708 * a pipe. The reverse isn't quite as easy, though. There are two possible
1709 * solutions for that:
1710 *
1711 * - memcpy() the data internally, at which point we might as well just
1712 * do a regular read() on the buffer anyway.
1713 * - Lots of nasty vm tricks, that are neither fast nor flexible (it
1714 * has restriction limitations on both ends of the pipe).
1715 *
1716 * Currently we punt and implement it as a normal copy, see pipe_to_user().
1717 *
1718 */
Heiko Carstens836f92a2009-01-14 14:14:33 +01001719SYSCALL_DEFINE4(vmsplice, int, fd, const struct iovec __user *, iov,
1720 unsigned long, nr_segs, unsigned int, flags)
Jens Axboe912d35f2006-04-26 10:59:21 +02001721{
1722 struct file *file;
1723 long error;
1724 int fput;
1725
Jens Axboe6a14b902007-06-14 13:08:55 +02001726 if (unlikely(nr_segs > UIO_MAXIOV))
1727 return -EINVAL;
1728 else if (unlikely(!nr_segs))
1729 return 0;
1730
Jens Axboe912d35f2006-04-26 10:59:21 +02001731 error = -EBADF;
1732 file = fget_light(fd, &fput);
1733 if (file) {
1734 if (file->f_mode & FMODE_WRITE)
Jens Axboe6a14b902007-06-14 13:08:55 +02001735 error = vmsplice_to_pipe(file, iov, nr_segs, flags);
1736 else if (file->f_mode & FMODE_READ)
1737 error = vmsplice_to_user(file, iov, nr_segs, flags);
Jens Axboe912d35f2006-04-26 10:59:21 +02001738
1739 fput_light(file, fput);
1740 }
1741
1742 return error;
1743}
1744
Heiko Carstens836f92a2009-01-14 14:14:33 +01001745SYSCALL_DEFINE6(splice, int, fd_in, loff_t __user *, off_in,
1746 int, fd_out, loff_t __user *, off_out,
1747 size_t, len, unsigned int, flags)
Jens Axboe5274f052006-03-30 15:15:30 +02001748{
1749 long error;
1750 struct file *in, *out;
1751 int fput_in, fput_out;
1752
1753 if (unlikely(!len))
1754 return 0;
1755
1756 error = -EBADF;
Ingo Molnar529565d2006-04-10 15:18:58 +02001757 in = fget_light(fd_in, &fput_in);
Jens Axboe5274f052006-03-30 15:15:30 +02001758 if (in) {
1759 if (in->f_mode & FMODE_READ) {
Ingo Molnar529565d2006-04-10 15:18:58 +02001760 out = fget_light(fd_out, &fput_out);
Jens Axboe5274f052006-03-30 15:15:30 +02001761 if (out) {
1762 if (out->f_mode & FMODE_WRITE)
Ingo Molnar529565d2006-04-10 15:18:58 +02001763 error = do_splice(in, off_in,
1764 out, off_out,
1765 len, flags);
Jens Axboe5274f052006-03-30 15:15:30 +02001766 fput_light(out, fput_out);
1767 }
1768 }
1769
1770 fput_light(in, fput_in);
1771 }
1772
1773 return error;
1774}
Jens Axboe70524492006-04-11 15:51:17 +02001775
1776/*
Jens Axboeaadd06e2006-07-10 11:00:01 +02001777 * Make sure there's data to read. Wait for input if we can, otherwise
1778 * return an appropriate error.
1779 */
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001780static int ipipe_prep(struct pipe_inode_info *pipe, unsigned int flags)
Jens Axboeaadd06e2006-07-10 11:00:01 +02001781{
1782 int ret;
1783
1784 /*
1785 * Check ->nrbufs without the inode lock first. This function
1786 * is speculative anyways, so missing one is ok.
1787 */
1788 if (pipe->nrbufs)
1789 return 0;
1790
1791 ret = 0;
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001792 pipe_lock(pipe);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001793
1794 while (!pipe->nrbufs) {
1795 if (signal_pending(current)) {
1796 ret = -ERESTARTSYS;
1797 break;
1798 }
1799 if (!pipe->writers)
1800 break;
1801 if (!pipe->waiting_writers) {
1802 if (flags & SPLICE_F_NONBLOCK) {
1803 ret = -EAGAIN;
1804 break;
1805 }
1806 }
1807 pipe_wait(pipe);
1808 }
1809
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001810 pipe_unlock(pipe);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001811 return ret;
1812}
1813
1814/*
1815 * Make sure there's writeable room. Wait for room if we can, otherwise
1816 * return an appropriate error.
1817 */
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001818static int opipe_prep(struct pipe_inode_info *pipe, unsigned int flags)
Jens Axboeaadd06e2006-07-10 11:00:01 +02001819{
1820 int ret;
1821
1822 /*
1823 * Check ->nrbufs without the inode lock first. This function
1824 * is speculative anyways, so missing one is ok.
1825 */
1826 if (pipe->nrbufs < PIPE_BUFFERS)
1827 return 0;
1828
1829 ret = 0;
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001830 pipe_lock(pipe);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001831
1832 while (pipe->nrbufs >= PIPE_BUFFERS) {
1833 if (!pipe->readers) {
1834 send_sig(SIGPIPE, current, 0);
1835 ret = -EPIPE;
1836 break;
1837 }
1838 if (flags & SPLICE_F_NONBLOCK) {
1839 ret = -EAGAIN;
1840 break;
1841 }
1842 if (signal_pending(current)) {
1843 ret = -ERESTARTSYS;
1844 break;
1845 }
1846 pipe->waiting_writers++;
1847 pipe_wait(pipe);
1848 pipe->waiting_writers--;
1849 }
1850
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001851 pipe_unlock(pipe);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001852 return ret;
1853}
1854
1855/*
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001856 * Splice contents of ipipe to opipe.
1857 */
1858static int splice_pipe_to_pipe(struct pipe_inode_info *ipipe,
1859 struct pipe_inode_info *opipe,
1860 size_t len, unsigned int flags)
1861{
1862 struct pipe_buffer *ibuf, *obuf;
1863 int ret = 0, nbuf;
1864 bool input_wakeup = false;
1865
1866
1867retry:
1868 ret = ipipe_prep(ipipe, flags);
1869 if (ret)
1870 return ret;
1871
1872 ret = opipe_prep(opipe, flags);
1873 if (ret)
1874 return ret;
1875
1876 /*
1877 * Potential ABBA deadlock, work around it by ordering lock
1878 * grabbing by pipe info address. Otherwise two different processes
1879 * could deadlock (one doing tee from A -> B, the other from B -> A).
1880 */
1881 pipe_double_lock(ipipe, opipe);
1882
1883 do {
1884 if (!opipe->readers) {
1885 send_sig(SIGPIPE, current, 0);
1886 if (!ret)
1887 ret = -EPIPE;
1888 break;
1889 }
1890
1891 if (!ipipe->nrbufs && !ipipe->writers)
1892 break;
1893
1894 /*
1895 * Cannot make any progress, because either the input
1896 * pipe is empty or the output pipe is full.
1897 */
1898 if (!ipipe->nrbufs || opipe->nrbufs >= PIPE_BUFFERS) {
1899 /* Already processed some buffers, break */
1900 if (ret)
1901 break;
1902
1903 if (flags & SPLICE_F_NONBLOCK) {
1904 ret = -EAGAIN;
1905 break;
1906 }
1907
1908 /*
1909 * We raced with another reader/writer and haven't
1910 * managed to process any buffers. A zero return
1911 * value means EOF, so retry instead.
1912 */
1913 pipe_unlock(ipipe);
1914 pipe_unlock(opipe);
1915 goto retry;
1916 }
1917
1918 ibuf = ipipe->bufs + ipipe->curbuf;
1919 nbuf = (opipe->curbuf + opipe->nrbufs) % PIPE_BUFFERS;
1920 obuf = opipe->bufs + nbuf;
1921
1922 if (len >= ibuf->len) {
1923 /*
1924 * Simply move the whole buffer from ipipe to opipe
1925 */
1926 *obuf = *ibuf;
1927 ibuf->ops = NULL;
1928 opipe->nrbufs++;
1929 ipipe->curbuf = (ipipe->curbuf + 1) % PIPE_BUFFERS;
1930 ipipe->nrbufs--;
1931 input_wakeup = true;
1932 } else {
1933 /*
1934 * Get a reference to this pipe buffer,
1935 * so we can copy the contents over.
1936 */
1937 ibuf->ops->get(ipipe, ibuf);
1938 *obuf = *ibuf;
1939
1940 /*
1941 * Don't inherit the gift flag, we need to
1942 * prevent multiple steals of this page.
1943 */
1944 obuf->flags &= ~PIPE_BUF_FLAG_GIFT;
1945
1946 obuf->len = len;
1947 opipe->nrbufs++;
1948 ibuf->offset += obuf->len;
1949 ibuf->len -= obuf->len;
1950 }
1951 ret += obuf->len;
1952 len -= obuf->len;
1953 } while (len);
1954
1955 pipe_unlock(ipipe);
1956 pipe_unlock(opipe);
1957
1958 /*
1959 * If we put data in the output pipe, wakeup any potential readers.
1960 */
1961 if (ret > 0) {
1962 smp_mb();
1963 if (waitqueue_active(&opipe->wait))
1964 wake_up_interruptible(&opipe->wait);
1965 kill_fasync(&opipe->fasync_readers, SIGIO, POLL_IN);
1966 }
1967 if (input_wakeup)
1968 wakeup_pipe_writers(ipipe);
1969
1970 return ret;
1971}
1972
1973/*
Jens Axboe70524492006-04-11 15:51:17 +02001974 * Link contents of ipipe to opipe.
1975 */
1976static int link_pipe(struct pipe_inode_info *ipipe,
1977 struct pipe_inode_info *opipe,
1978 size_t len, unsigned int flags)
1979{
1980 struct pipe_buffer *ibuf, *obuf;
Jens Axboeaadd06e2006-07-10 11:00:01 +02001981 int ret = 0, i = 0, nbuf;
Jens Axboe70524492006-04-11 15:51:17 +02001982
1983 /*
1984 * Potential ABBA deadlock, work around it by ordering lock
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001985 * grabbing by pipe info address. Otherwise two different processes
Jens Axboe70524492006-04-11 15:51:17 +02001986 * could deadlock (one doing tee from A -> B, the other from B -> A).
1987 */
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001988 pipe_double_lock(ipipe, opipe);
Jens Axboe70524492006-04-11 15:51:17 +02001989
Jens Axboeaadd06e2006-07-10 11:00:01 +02001990 do {
Jens Axboe70524492006-04-11 15:51:17 +02001991 if (!opipe->readers) {
1992 send_sig(SIGPIPE, current, 0);
1993 if (!ret)
1994 ret = -EPIPE;
1995 break;
1996 }
Jens Axboe70524492006-04-11 15:51:17 +02001997
1998 /*
Jens Axboeaadd06e2006-07-10 11:00:01 +02001999 * If we have iterated all input buffers or ran out of
2000 * output room, break.
Jens Axboe70524492006-04-11 15:51:17 +02002001 */
Jens Axboeaadd06e2006-07-10 11:00:01 +02002002 if (i >= ipipe->nrbufs || opipe->nrbufs >= PIPE_BUFFERS)
Jens Axboe70524492006-04-11 15:51:17 +02002003 break;
Jens Axboeaadd06e2006-07-10 11:00:01 +02002004
2005 ibuf = ipipe->bufs + ((ipipe->curbuf + i) & (PIPE_BUFFERS - 1));
2006 nbuf = (opipe->curbuf + opipe->nrbufs) & (PIPE_BUFFERS - 1);
2007
Jens Axboe2a272502006-04-19 15:56:40 +02002008 /*
Jens Axboeaadd06e2006-07-10 11:00:01 +02002009 * Get a reference to this pipe buffer,
2010 * so we can copy the contents over.
Jens Axboe2a272502006-04-19 15:56:40 +02002011 */
Jens Axboeaadd06e2006-07-10 11:00:01 +02002012 ibuf->ops->get(ipipe, ibuf);
Jens Axboe70524492006-04-11 15:51:17 +02002013
Jens Axboeaadd06e2006-07-10 11:00:01 +02002014 obuf = opipe->bufs + nbuf;
2015 *obuf = *ibuf;
Jens Axboe70524492006-04-11 15:51:17 +02002016
Jens Axboeaadd06e2006-07-10 11:00:01 +02002017 /*
2018 * Don't inherit the gift flag, we need to
2019 * prevent multiple steals of this page.
2020 */
2021 obuf->flags &= ~PIPE_BUF_FLAG_GIFT;
2022
2023 if (obuf->len > len)
2024 obuf->len = len;
2025
2026 opipe->nrbufs++;
2027 ret += obuf->len;
2028 len -= obuf->len;
2029 i++;
2030 } while (len);
Jens Axboe70524492006-04-11 15:51:17 +02002031
Jens Axboe02cf01a2008-02-20 10:34:51 +01002032 /*
2033 * return EAGAIN if we have the potential of some data in the
2034 * future, otherwise just return 0
2035 */
2036 if (!ret && ipipe->waiting_writers && (flags & SPLICE_F_NONBLOCK))
2037 ret = -EAGAIN;
2038
Miklos Szeredi61e0d472009-04-14 19:48:41 +02002039 pipe_unlock(ipipe);
2040 pipe_unlock(opipe);
Jens Axboe70524492006-04-11 15:51:17 +02002041
Jens Axboeaadd06e2006-07-10 11:00:01 +02002042 /*
2043 * If we put data in the output pipe, wakeup any potential readers.
2044 */
2045 if (ret > 0) {
Jens Axboe70524492006-04-11 15:51:17 +02002046 smp_mb();
2047 if (waitqueue_active(&opipe->wait))
2048 wake_up_interruptible(&opipe->wait);
2049 kill_fasync(&opipe->fasync_readers, SIGIO, POLL_IN);
2050 }
2051
2052 return ret;
2053}
2054
2055/*
2056 * This is a tee(1) implementation that works on pipes. It doesn't copy
2057 * any data, it simply references the 'in' pages on the 'out' pipe.
2058 * The 'flags' used are the SPLICE_F_* variants, currently the only
2059 * applicable one is SPLICE_F_NONBLOCK.
2060 */
2061static long do_tee(struct file *in, struct file *out, size_t len,
2062 unsigned int flags)
2063{
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08002064 struct pipe_inode_info *ipipe = pipe_info(in->f_path.dentry->d_inode);
2065 struct pipe_inode_info *opipe = pipe_info(out->f_path.dentry->d_inode);
Jens Axboeaadd06e2006-07-10 11:00:01 +02002066 int ret = -EINVAL;
Jens Axboe70524492006-04-11 15:51:17 +02002067
2068 /*
Jens Axboeaadd06e2006-07-10 11:00:01 +02002069 * Duplicate the contents of ipipe to opipe without actually
2070 * copying the data.
Jens Axboe70524492006-04-11 15:51:17 +02002071 */
Jens Axboeaadd06e2006-07-10 11:00:01 +02002072 if (ipipe && opipe && ipipe != opipe) {
2073 /*
2074 * Keep going, unless we encounter an error. The ipipe/opipe
2075 * ordering doesn't really matter.
2076 */
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02002077 ret = ipipe_prep(ipipe, flags);
Jens Axboeaadd06e2006-07-10 11:00:01 +02002078 if (!ret) {
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02002079 ret = opipe_prep(opipe, flags);
Jens Axboe02cf01a2008-02-20 10:34:51 +01002080 if (!ret)
Jens Axboeaadd06e2006-07-10 11:00:01 +02002081 ret = link_pipe(ipipe, opipe, len, flags);
Jens Axboeaadd06e2006-07-10 11:00:01 +02002082 }
2083 }
Jens Axboe70524492006-04-11 15:51:17 +02002084
Jens Axboeaadd06e2006-07-10 11:00:01 +02002085 return ret;
Jens Axboe70524492006-04-11 15:51:17 +02002086}
2087
Heiko Carstens836f92a2009-01-14 14:14:33 +01002088SYSCALL_DEFINE4(tee, int, fdin, int, fdout, size_t, len, unsigned int, flags)
Jens Axboe70524492006-04-11 15:51:17 +02002089{
2090 struct file *in;
2091 int error, fput_in;
2092
2093 if (unlikely(!len))
2094 return 0;
2095
2096 error = -EBADF;
2097 in = fget_light(fdin, &fput_in);
2098 if (in) {
2099 if (in->f_mode & FMODE_READ) {
2100 int fput_out;
2101 struct file *out = fget_light(fdout, &fput_out);
2102
2103 if (out) {
2104 if (out->f_mode & FMODE_WRITE)
2105 error = do_tee(in, out, len, flags);
2106 fput_light(out, fput_out);
2107 }
2108 }
2109 fput_light(in, fput_in);
2110 }
2111
2112 return error;
2113}