blob: 73766d24f97b9d41f48ba2cadf36d4cd7ae3d60b [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 Axboecac36bb02007-06-14 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 Axboecac36bb02007-06-14 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 Axboecac36bb02007-06-14 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 Szeredib2858d72009-05-19 11:37:46 +0200538static ssize_t kernel_write(struct file *file, const char *buf, size_t count,
539 loff_t pos)
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +0200540{
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() */
Miklos Szeredib2858d72009-05-19 11:37:46 +0200547 res = vfs_write(file, (const char __user *)buf, count, &pos);
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +0200548 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
Jens Axboe4f231222009-05-13 08:35:35 +0200583 page = alloc_page(GFP_USER);
Miklos Szeredi68181732009-05-07 15:37:36 +0200584 error = -ENOMEM;
585 if (!page)
586 goto err;
587
588 this_len = min_t(size_t, len, PAGE_CACHE_SIZE - offset);
Jens Axboe4f231222009-05-13 08:35:35 +0200589 vec[i].iov_base = (void __user *) page_address(page);
Miklos Szeredi68181732009-05-07 15:37:36 +0200590 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);
Andrew Morton77f6bf52009-05-14 09:49:44 +0200598 if (res < 0) {
599 error = res;
Miklos Szeredi68181732009-05-07 15:37:36 +0200600 goto err;
Andrew Morton77f6bf52009-05-14 09:49:44 +0200601 }
Miklos Szeredi68181732009-05-07 15:37:36 +0200602
603 error = 0;
604 if (!res)
605 goto err;
606
607 nr_freed = 0;
608 for (i = 0; i < spd.nr_pages; i++) {
Miklos Szeredi68181732009-05-07 15:37:36 +0200609 this_len = min_t(size_t, vec[i].iov_len, res);
610 partial[i].offset = 0;
611 partial[i].len = this_len;
612 if (!this_len) {
613 __free_page(pages[i]);
614 pages[i] = NULL;
615 nr_freed++;
616 }
617 res -= this_len;
618 }
619 spd.nr_pages -= nr_freed;
620
621 res = splice_to_pipe(pipe, &spd);
622 if (res > 0)
623 *ppos += res;
624
625 return res;
626
627err:
Jens Axboe4f231222009-05-13 08:35:35 +0200628 for (i = 0; i < spd.nr_pages; i++)
Miklos Szeredi68181732009-05-07 15:37:36 +0200629 __free_page(pages[i]);
Jens Axboe4f231222009-05-13 08:35:35 +0200630
Miklos Szeredi68181732009-05-07 15:37:36 +0200631 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 Axboecac36bb02007-06-14 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 Axboecac36bb02007-06-14 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 Fasheh40bee44e2007-03-21 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 Szeredib2858d72009-05-19 11:37:46 +02001006static int write_pipe_buf(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
1007 struct splice_desc *sd)
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +02001008{
Miklos Szeredib2858d72009-05-19 11:37:46 +02001009 int ret;
1010 void *data;
1011
1012 ret = buf->ops->confirm(pipe, buf);
1013 if (ret)
1014 return ret;
1015
1016 data = buf->ops->map(pipe, buf, 0);
1017 ret = kernel_write(sd->u.file, data + buf->offset, sd->len, sd->pos);
1018 buf->ops->unmap(pipe, buf, data);
1019
1020 return ret;
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +02001021}
1022
1023static ssize_t default_file_splice_write(struct pipe_inode_info *pipe,
1024 struct file *out, loff_t *ppos,
1025 size_t len, unsigned int flags)
1026{
Miklos Szeredib2858d72009-05-19 11:37:46 +02001027 ssize_t ret;
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +02001028
Miklos Szeredib2858d72009-05-19 11:37:46 +02001029 ret = splice_from_pipe(pipe, out, ppos, len, flags, write_pipe_buf);
1030 if (ret > 0)
1031 *ppos += ret;
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +02001032
Miklos Szeredib2858d72009-05-19 11:37:46 +02001033 return ret;
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +02001034}
1035
Jens Axboe83f91352006-04-02 23:05:09 +02001036/**
1037 * generic_splice_sendpage - splice data from a pipe to a socket
Jens Axboe932cc6d2007-06-21 13:10:21 +02001038 * @pipe: pipe to splice from
Jens Axboe83f91352006-04-02 23:05:09 +02001039 * @out: socket to write to
Jens Axboe932cc6d2007-06-21 13:10:21 +02001040 * @ppos: position in @out
Jens Axboe83f91352006-04-02 23:05:09 +02001041 * @len: number of bytes to splice
1042 * @flags: splice modifier flags
1043 *
Jens Axboe932cc6d2007-06-21 13:10:21 +02001044 * Description:
1045 * Will send @len bytes from the pipe to a network socket. No data copying
1046 * is involved.
Jens Axboe83f91352006-04-02 23:05:09 +02001047 *
1048 */
Ingo Molnar3a326a22006-04-10 15:18:35 +02001049ssize_t generic_splice_sendpage(struct pipe_inode_info *pipe, struct file *out,
Jens Axboecbb7e572006-04-11 14:57:50 +02001050 loff_t *ppos, size_t len, unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +02001051{
Jens Axboe00522fb2006-04-26 14:39:29 +02001052 return splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_sendpage);
Jens Axboe5274f052006-03-30 15:15:30 +02001053}
1054
Jens Axboe059a8f32006-04-02 23:06:05 +02001055EXPORT_SYMBOL(generic_splice_sendpage);
Jeff Garzika0f06782006-03-30 23:06:13 -05001056
Jens Axboe83f91352006-04-02 23:05:09 +02001057/*
1058 * Attempt to initiate a splice from pipe to file.
1059 */
Ingo Molnar3a326a22006-04-10 15:18:35 +02001060static long do_splice_from(struct pipe_inode_info *pipe, struct file *out,
Jens Axboecbb7e572006-04-11 14:57:50 +02001061 loff_t *ppos, size_t len, unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +02001062{
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +02001063 ssize_t (*splice_write)(struct pipe_inode_info *, struct file *,
1064 loff_t *, size_t, unsigned int);
Jens Axboe5274f052006-03-30 15:15:30 +02001065 int ret;
1066
Jens Axboe49570e92006-04-11 13:56:09 +02001067 if (unlikely(!(out->f_mode & FMODE_WRITE)))
Jens Axboe5274f052006-03-30 15:15:30 +02001068 return -EBADF;
1069
Linus Torvaldsefc968d2008-10-09 14:04:54 -07001070 if (unlikely(out->f_flags & O_APPEND))
1071 return -EINVAL;
1072
Jens Axboecbb7e572006-04-11 14:57:50 +02001073 ret = rw_verify_area(WRITE, out, ppos, len);
Jens Axboe5274f052006-03-30 15:15:30 +02001074 if (unlikely(ret < 0))
1075 return ret;
1076
Miklos Szeredi0b0a47f2009-05-07 15:37:37 +02001077 splice_write = out->f_op->splice_write;
1078 if (!splice_write)
1079 splice_write = default_file_splice_write;
1080
1081 return splice_write(pipe, out, ppos, len, flags);
Jens Axboe5274f052006-03-30 15:15:30 +02001082}
1083
Jens Axboe83f91352006-04-02 23:05:09 +02001084/*
1085 * Attempt to initiate a splice from a file to a pipe.
1086 */
Jens Axboecbb7e572006-04-11 14:57:50 +02001087static long do_splice_to(struct file *in, loff_t *ppos,
1088 struct pipe_inode_info *pipe, size_t len,
1089 unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +02001090{
Miklos Szeredi68181732009-05-07 15:37:36 +02001091 ssize_t (*splice_read)(struct file *, loff_t *,
1092 struct pipe_inode_info *, size_t, unsigned int);
Jens Axboe5274f052006-03-30 15:15:30 +02001093 int ret;
1094
Jens Axboe49570e92006-04-11 13:56:09 +02001095 if (unlikely(!(in->f_mode & FMODE_READ)))
Jens Axboe5274f052006-03-30 15:15:30 +02001096 return -EBADF;
1097
Jens Axboecbb7e572006-04-11 14:57:50 +02001098 ret = rw_verify_area(READ, in, ppos, len);
Jens Axboe5274f052006-03-30 15:15:30 +02001099 if (unlikely(ret < 0))
1100 return ret;
1101
Miklos Szeredi68181732009-05-07 15:37:36 +02001102 splice_read = in->f_op->splice_read;
1103 if (!splice_read)
1104 splice_read = default_file_splice_read;
1105
1106 return splice_read(in, ppos, pipe, len, flags);
Jens Axboe5274f052006-03-30 15:15:30 +02001107}
1108
Jens Axboe932cc6d2007-06-21 13:10:21 +02001109/**
1110 * splice_direct_to_actor - splices data directly between two non-pipes
1111 * @in: file to splice from
1112 * @sd: actor information on where to splice to
1113 * @actor: handles the data splicing
1114 *
1115 * Description:
1116 * This is a special case helper to splice directly between two
1117 * points, without requiring an explicit pipe. Internally an allocated
Randy Dunlap79685b82007-07-27 08:08:51 +02001118 * pipe is cached in the process, and reused during the lifetime of
Jens Axboe932cc6d2007-06-21 13:10:21 +02001119 * that process.
1120 *
Jens Axboec66ab6f2007-06-12 21:17:17 +02001121 */
1122ssize_t splice_direct_to_actor(struct file *in, struct splice_desc *sd,
1123 splice_direct_actor *actor)
Jens Axboeb92ce552006-04-11 13:52:07 +02001124{
1125 struct pipe_inode_info *pipe;
1126 long ret, bytes;
1127 umode_t i_mode;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001128 size_t len;
1129 int i, flags;
Jens Axboeb92ce552006-04-11 13:52:07 +02001130
1131 /*
1132 * We require the input being a regular file, as we don't want to
1133 * randomly drop data for eg socket -> socket splicing. Use the
1134 * piped splicing for that!
1135 */
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001136 i_mode = in->f_path.dentry->d_inode->i_mode;
Jens Axboeb92ce552006-04-11 13:52:07 +02001137 if (unlikely(!S_ISREG(i_mode) && !S_ISBLK(i_mode)))
1138 return -EINVAL;
1139
1140 /*
1141 * neither in nor out is a pipe, setup an internal pipe attached to
1142 * 'out' and transfer the wanted data from 'in' to 'out' through that
1143 */
1144 pipe = current->splice_pipe;
Jens Axboe49570e92006-04-11 13:56:09 +02001145 if (unlikely(!pipe)) {
Jens Axboeb92ce552006-04-11 13:52:07 +02001146 pipe = alloc_pipe_info(NULL);
1147 if (!pipe)
1148 return -ENOMEM;
1149
1150 /*
1151 * We don't have an immediate reader, but we'll read the stuff
Jens Axboe00522fb2006-04-26 14:39:29 +02001152 * out of the pipe right after the splice_to_pipe(). So set
Jens Axboeb92ce552006-04-11 13:52:07 +02001153 * PIPE_READERS appropriately.
1154 */
1155 pipe->readers = 1;
1156
1157 current->splice_pipe = pipe;
1158 }
1159
1160 /*
Ingo Molnar73d62d82006-04-11 13:57:21 +02001161 * Do the splice.
Jens Axboeb92ce552006-04-11 13:52:07 +02001162 */
1163 ret = 0;
1164 bytes = 0;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001165 len = sd->total_len;
1166 flags = sd->flags;
1167
1168 /*
1169 * Don't block on output, we have to drain the direct pipe.
1170 */
1171 sd->flags &= ~SPLICE_F_NONBLOCK;
Jens Axboeb92ce552006-04-11 13:52:07 +02001172
1173 while (len) {
Jens Axboe51a92c02007-07-13 14:11:43 +02001174 size_t read_len;
Tom Zanussia82c53a2008-05-09 13:28:36 +02001175 loff_t pos = sd->pos, prev_pos = pos;
Jens Axboeb92ce552006-04-11 13:52:07 +02001176
Jens Axboebcd4f3a2007-07-16 14:41:49 +02001177 ret = do_splice_to(in, &pos, pipe, len, flags);
Jens Axboe51a92c02007-07-13 14:11:43 +02001178 if (unlikely(ret <= 0))
Jens Axboeb92ce552006-04-11 13:52:07 +02001179 goto out_release;
1180
1181 read_len = ret;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001182 sd->total_len = read_len;
Jens Axboeb92ce552006-04-11 13:52:07 +02001183
1184 /*
1185 * NOTE: nonblocking mode only applies to the input. We
1186 * must not do the output in nonblocking mode as then we
1187 * could get stuck data in the internal pipe:
1188 */
Jens Axboec66ab6f2007-06-12 21:17:17 +02001189 ret = actor(pipe, sd);
Tom Zanussia82c53a2008-05-09 13:28:36 +02001190 if (unlikely(ret <= 0)) {
1191 sd->pos = prev_pos;
Jens Axboeb92ce552006-04-11 13:52:07 +02001192 goto out_release;
Tom Zanussia82c53a2008-05-09 13:28:36 +02001193 }
Jens Axboeb92ce552006-04-11 13:52:07 +02001194
1195 bytes += ret;
1196 len -= ret;
Jens Axboebcd4f3a2007-07-16 14:41:49 +02001197 sd->pos = pos;
Jens Axboeb92ce552006-04-11 13:52:07 +02001198
Tom Zanussia82c53a2008-05-09 13:28:36 +02001199 if (ret < read_len) {
1200 sd->pos = prev_pos + ret;
Jens Axboe51a92c02007-07-13 14:11:43 +02001201 goto out_release;
Tom Zanussia82c53a2008-05-09 13:28:36 +02001202 }
Jens Axboeb92ce552006-04-11 13:52:07 +02001203 }
1204
Jens Axboe9e971982008-01-29 21:05:57 +01001205done:
Jens Axboeb92ce552006-04-11 13:52:07 +02001206 pipe->nrbufs = pipe->curbuf = 0;
Jens Axboe80848702008-01-30 12:24:48 +01001207 file_accessed(in);
Jens Axboeb92ce552006-04-11 13:52:07 +02001208 return bytes;
1209
1210out_release:
1211 /*
1212 * If we did an incomplete transfer we must release
1213 * the pipe buffers in question:
1214 */
1215 for (i = 0; i < PIPE_BUFFERS; i++) {
1216 struct pipe_buffer *buf = pipe->bufs + i;
1217
1218 if (buf->ops) {
1219 buf->ops->release(pipe, buf);
1220 buf->ops = NULL;
1221 }
1222 }
Jens Axboeb92ce552006-04-11 13:52:07 +02001223
Jens Axboe9e971982008-01-29 21:05:57 +01001224 if (!bytes)
1225 bytes = ret;
Jens Axboeb92ce552006-04-11 13:52:07 +02001226
Jens Axboe9e971982008-01-29 21:05:57 +01001227 goto done;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001228}
1229EXPORT_SYMBOL(splice_direct_to_actor);
1230
1231static int direct_splice_actor(struct pipe_inode_info *pipe,
1232 struct splice_desc *sd)
1233{
Jens Axboe6a14b902007-06-14 13:08:55 +02001234 struct file *file = sd->u.file;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001235
1236 return do_splice_from(pipe, file, &sd->pos, sd->total_len, sd->flags);
1237}
1238
Jens Axboe932cc6d2007-06-21 13:10:21 +02001239/**
1240 * do_splice_direct - splices data directly between two files
1241 * @in: file to splice from
1242 * @ppos: input file offset
1243 * @out: file to splice to
1244 * @len: number of bytes to splice
1245 * @flags: splice modifier flags
1246 *
1247 * Description:
1248 * For use by do_sendfile(). splice can easily emulate sendfile, but
1249 * doing it in the application would incur an extra system call
1250 * (splice in + splice out, as compared to just sendfile()). So this helper
1251 * can splice directly through a process-private pipe.
1252 *
1253 */
Jens Axboec66ab6f2007-06-12 21:17:17 +02001254long do_splice_direct(struct file *in, loff_t *ppos, struct file *out,
1255 size_t len, unsigned int flags)
1256{
1257 struct splice_desc sd = {
1258 .len = len,
1259 .total_len = len,
1260 .flags = flags,
1261 .pos = *ppos,
Jens Axboe6a14b902007-06-14 13:08:55 +02001262 .u.file = out,
Jens Axboec66ab6f2007-06-12 21:17:17 +02001263 };
Jens Axboe51a92c02007-07-13 14:11:43 +02001264 long ret;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001265
1266 ret = splice_direct_to_actor(in, &sd, direct_splice_actor);
Jens Axboe51a92c02007-07-13 14:11:43 +02001267 if (ret > 0)
Tom Zanussia82c53a2008-05-09 13:28:36 +02001268 *ppos = sd.pos;
Jens Axboe51a92c02007-07-13 14:11:43 +02001269
Jens Axboec66ab6f2007-06-12 21:17:17 +02001270 return ret;
Jens Axboeb92ce552006-04-11 13:52:07 +02001271}
1272
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001273static int splice_pipe_to_pipe(struct pipe_inode_info *ipipe,
1274 struct pipe_inode_info *opipe,
1275 size_t len, unsigned int flags);
Jens Axboe83f91352006-04-02 23:05:09 +02001276/*
Jens Axboeddac0d32006-11-04 12:49:32 +01001277 * After the inode slimming patch, i_pipe/i_bdev/i_cdev share the same
1278 * location, so checking ->i_pipe is not enough to verify that this is a
1279 * pipe.
1280 */
1281static inline struct pipe_inode_info *pipe_info(struct inode *inode)
1282{
1283 if (S_ISFIFO(inode->i_mode))
1284 return inode->i_pipe;
1285
1286 return NULL;
1287}
1288
1289/*
Jens Axboe83f91352006-04-02 23:05:09 +02001290 * Determine where to splice to/from.
1291 */
Ingo Molnar529565d2006-04-10 15:18:58 +02001292static long do_splice(struct file *in, loff_t __user *off_in,
1293 struct file *out, loff_t __user *off_out,
1294 size_t len, unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +02001295{
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001296 struct pipe_inode_info *ipipe;
1297 struct pipe_inode_info *opipe;
Jens Axboecbb7e572006-04-11 14:57:50 +02001298 loff_t offset, *off;
Jens Axboea4514eb2006-04-19 15:57:05 +02001299 long ret;
Jens Axboe5274f052006-03-30 15:15:30 +02001300
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001301 ipipe = pipe_info(in->f_path.dentry->d_inode);
1302 opipe = pipe_info(out->f_path.dentry->d_inode);
1303
1304 if (ipipe && opipe) {
1305 if (off_in || off_out)
1306 return -ESPIPE;
1307
1308 if (!(in->f_mode & FMODE_READ))
1309 return -EBADF;
1310
1311 if (!(out->f_mode & FMODE_WRITE))
1312 return -EBADF;
1313
1314 /* Splicing to self would be fun, but... */
1315 if (ipipe == opipe)
1316 return -EINVAL;
1317
1318 return splice_pipe_to_pipe(ipipe, opipe, len, flags);
1319 }
1320
1321 if (ipipe) {
Ingo Molnar529565d2006-04-10 15:18:58 +02001322 if (off_in)
1323 return -ESPIPE;
Jens Axboeb92ce552006-04-11 13:52:07 +02001324 if (off_out) {
1325 if (out->f_op->llseek == no_llseek)
1326 return -EINVAL;
Jens Axboecbb7e572006-04-11 14:57:50 +02001327 if (copy_from_user(&offset, off_out, sizeof(loff_t)))
Jens Axboeb92ce552006-04-11 13:52:07 +02001328 return -EFAULT;
Jens Axboecbb7e572006-04-11 14:57:50 +02001329 off = &offset;
1330 } else
1331 off = &out->f_pos;
Ingo Molnar529565d2006-04-10 15:18:58 +02001332
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001333 ret = do_splice_from(ipipe, out, off, len, flags);
Jens Axboea4514eb2006-04-19 15:57:05 +02001334
1335 if (off_out && copy_to_user(off_out, off, sizeof(loff_t)))
1336 ret = -EFAULT;
1337
1338 return ret;
Ingo Molnar529565d2006-04-10 15:18:58 +02001339 }
Jens Axboe5274f052006-03-30 15:15:30 +02001340
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001341 if (opipe) {
Ingo Molnar529565d2006-04-10 15:18:58 +02001342 if (off_out)
1343 return -ESPIPE;
Jens Axboeb92ce552006-04-11 13:52:07 +02001344 if (off_in) {
1345 if (in->f_op->llseek == no_llseek)
1346 return -EINVAL;
Jens Axboecbb7e572006-04-11 14:57:50 +02001347 if (copy_from_user(&offset, off_in, sizeof(loff_t)))
Jens Axboeb92ce552006-04-11 13:52:07 +02001348 return -EFAULT;
Jens Axboecbb7e572006-04-11 14:57:50 +02001349 off = &offset;
1350 } else
1351 off = &in->f_pos;
Ingo Molnar529565d2006-04-10 15:18:58 +02001352
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001353 ret = do_splice_to(in, off, opipe, len, flags);
Jens Axboea4514eb2006-04-19 15:57:05 +02001354
1355 if (off_in && copy_to_user(off_in, off, sizeof(loff_t)))
1356 ret = -EFAULT;
1357
1358 return ret;
Ingo Molnar529565d2006-04-10 15:18:58 +02001359 }
Jens Axboe5274f052006-03-30 15:15:30 +02001360
1361 return -EINVAL;
1362}
1363
Jens Axboe912d35f2006-04-26 10:59:21 +02001364/*
1365 * Map an iov into an array of pages and offset/length tupples. With the
1366 * partial_page structure, we can map several non-contiguous ranges into
1367 * our ones pages[] map instead of splitting that operation into pieces.
1368 * Could easily be exported as a generic helper for other users, in which
1369 * case one would probably want to add a 'max_nr_pages' parameter as well.
1370 */
1371static int get_iovec_page_array(const struct iovec __user *iov,
1372 unsigned int nr_vecs, struct page **pages,
Jens Axboe7afa6fd2006-05-01 20:02:33 +02001373 struct partial_page *partial, int aligned)
Jens Axboe912d35f2006-04-26 10:59:21 +02001374{
1375 int buffers = 0, error = 0;
1376
Jens Axboe912d35f2006-04-26 10:59:21 +02001377 while (nr_vecs) {
1378 unsigned long off, npages;
Linus Torvalds75723952007-10-01 13:17:28 -07001379 struct iovec entry;
Jens Axboe912d35f2006-04-26 10:59:21 +02001380 void __user *base;
1381 size_t len;
1382 int i;
1383
Linus Torvalds75723952007-10-01 13:17:28 -07001384 error = -EFAULT;
Nick Pigginbc40d732008-07-25 19:45:26 -07001385 if (copy_from_user(&entry, iov, sizeof(entry)))
Jens Axboe912d35f2006-04-26 10:59:21 +02001386 break;
Linus Torvalds75723952007-10-01 13:17:28 -07001387
1388 base = entry.iov_base;
1389 len = entry.iov_len;
Jens Axboe912d35f2006-04-26 10:59:21 +02001390
1391 /*
1392 * Sanity check this iovec. 0 read succeeds.
1393 */
Linus Torvalds75723952007-10-01 13:17:28 -07001394 error = 0;
Jens Axboe912d35f2006-04-26 10:59:21 +02001395 if (unlikely(!len))
1396 break;
1397 error = -EFAULT;
Bastian Blank712a30e2008-02-10 16:47:57 +02001398 if (!access_ok(VERIFY_READ, base, len))
Jens Axboe912d35f2006-04-26 10:59:21 +02001399 break;
1400
1401 /*
1402 * Get this base offset and number of pages, then map
1403 * in the user pages.
1404 */
1405 off = (unsigned long) base & ~PAGE_MASK;
Jens Axboe7afa6fd2006-05-01 20:02:33 +02001406
1407 /*
1408 * If asked for alignment, the offset must be zero and the
1409 * length a multiple of the PAGE_SIZE.
1410 */
1411 error = -EINVAL;
1412 if (aligned && (off || len & ~PAGE_MASK))
1413 break;
1414
Jens Axboe912d35f2006-04-26 10:59:21 +02001415 npages = (off + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1416 if (npages > PIPE_BUFFERS - buffers)
1417 npages = PIPE_BUFFERS - buffers;
1418
Nick Pigginbc40d732008-07-25 19:45:26 -07001419 error = get_user_pages_fast((unsigned long)base, npages,
1420 0, &pages[buffers]);
Jens Axboe912d35f2006-04-26 10:59:21 +02001421
1422 if (unlikely(error <= 0))
1423 break;
1424
1425 /*
1426 * Fill this contiguous range into the partial page map.
1427 */
1428 for (i = 0; i < error; i++) {
Jens Axboe75914892006-05-02 12:57:18 +02001429 const int plen = min_t(size_t, len, PAGE_SIZE - off);
Jens Axboe912d35f2006-04-26 10:59:21 +02001430
1431 partial[buffers].offset = off;
1432 partial[buffers].len = plen;
1433
1434 off = 0;
1435 len -= plen;
1436 buffers++;
1437 }
1438
1439 /*
1440 * We didn't complete this iov, stop here since it probably
1441 * means we have to move some of this into a pipe to
1442 * be able to continue.
1443 */
1444 if (len)
1445 break;
1446
1447 /*
1448 * Don't continue if we mapped fewer pages than we asked for,
1449 * or if we mapped the max number of pages that we have
1450 * room for.
1451 */
1452 if (error < npages || buffers == PIPE_BUFFERS)
1453 break;
1454
1455 nr_vecs--;
1456 iov++;
1457 }
1458
Jens Axboe912d35f2006-04-26 10:59:21 +02001459 if (buffers)
1460 return buffers;
1461
1462 return error;
1463}
1464
Jens Axboe6a14b902007-06-14 13:08:55 +02001465static int pipe_to_user(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
1466 struct splice_desc *sd)
1467{
1468 char *src;
1469 int ret;
1470
Jens Axboecac36bb02007-06-14 13:10:48 +02001471 ret = buf->ops->confirm(pipe, buf);
Jens Axboe6a14b902007-06-14 13:08:55 +02001472 if (unlikely(ret))
1473 return ret;
1474
1475 /*
1476 * See if we can use the atomic maps, by prefaulting in the
1477 * pages and doing an atomic copy
1478 */
1479 if (!fault_in_pages_writeable(sd->u.userptr, sd->len)) {
1480 src = buf->ops->map(pipe, buf, 1);
1481 ret = __copy_to_user_inatomic(sd->u.userptr, src + buf->offset,
1482 sd->len);
1483 buf->ops->unmap(pipe, buf, src);
1484 if (!ret) {
1485 ret = sd->len;
1486 goto out;
1487 }
1488 }
1489
1490 /*
1491 * No dice, use slow non-atomic map and copy
1492 */
1493 src = buf->ops->map(pipe, buf, 0);
1494
1495 ret = sd->len;
1496 if (copy_to_user(sd->u.userptr, src + buf->offset, sd->len))
1497 ret = -EFAULT;
1498
Jens Axboe6866bef2007-10-16 10:01:29 +02001499 buf->ops->unmap(pipe, buf, src);
Jens Axboe6a14b902007-06-14 13:08:55 +02001500out:
1501 if (ret > 0)
1502 sd->u.userptr += ret;
Jens Axboe6a14b902007-06-14 13:08:55 +02001503 return ret;
1504}
1505
1506/*
1507 * For lack of a better implementation, implement vmsplice() to userspace
1508 * as a simple copy of the pipes pages to the user iov.
1509 */
1510static long vmsplice_to_user(struct file *file, const struct iovec __user *iov,
1511 unsigned long nr_segs, unsigned int flags)
1512{
1513 struct pipe_inode_info *pipe;
1514 struct splice_desc sd;
1515 ssize_t size;
1516 int error;
1517 long ret;
1518
1519 pipe = pipe_info(file->f_path.dentry->d_inode);
1520 if (!pipe)
1521 return -EBADF;
1522
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001523 pipe_lock(pipe);
Jens Axboe6a14b902007-06-14 13:08:55 +02001524
1525 error = ret = 0;
1526 while (nr_segs) {
1527 void __user *base;
1528 size_t len;
1529
1530 /*
1531 * Get user address base and length for this iovec.
1532 */
1533 error = get_user(base, &iov->iov_base);
1534 if (unlikely(error))
1535 break;
1536 error = get_user(len, &iov->iov_len);
1537 if (unlikely(error))
1538 break;
1539
1540 /*
1541 * Sanity check this iovec. 0 read succeeds.
1542 */
1543 if (unlikely(!len))
1544 break;
1545 if (unlikely(!base)) {
1546 error = -EFAULT;
1547 break;
1548 }
1549
Jens Axboe88119302008-02-08 08:49:14 -08001550 if (unlikely(!access_ok(VERIFY_WRITE, base, len))) {
1551 error = -EFAULT;
1552 break;
1553 }
1554
Jens Axboe6a14b902007-06-14 13:08:55 +02001555 sd.len = 0;
1556 sd.total_len = len;
1557 sd.flags = flags;
1558 sd.u.userptr = base;
1559 sd.pos = 0;
1560
1561 size = __splice_from_pipe(pipe, &sd, pipe_to_user);
1562 if (size < 0) {
1563 if (!ret)
1564 ret = size;
1565
1566 break;
1567 }
1568
1569 ret += size;
1570
1571 if (size < len)
1572 break;
1573
1574 nr_segs--;
1575 iov++;
1576 }
1577
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001578 pipe_unlock(pipe);
Jens Axboe6a14b902007-06-14 13:08:55 +02001579
1580 if (!ret)
1581 ret = error;
1582
1583 return ret;
1584}
1585
Jens Axboe912d35f2006-04-26 10:59:21 +02001586/*
1587 * vmsplice splices a user address range into a pipe. It can be thought of
1588 * as splice-from-memory, where the regular splice is splice-from-file (or
1589 * to file). In both cases the output is a pipe, naturally.
Jens Axboe912d35f2006-04-26 10:59:21 +02001590 */
Jens Axboe6a14b902007-06-14 13:08:55 +02001591static long vmsplice_to_pipe(struct file *file, const struct iovec __user *iov,
1592 unsigned long nr_segs, unsigned int flags)
Jens Axboe912d35f2006-04-26 10:59:21 +02001593{
Jens Axboeddac0d32006-11-04 12:49:32 +01001594 struct pipe_inode_info *pipe;
Jens Axboe912d35f2006-04-26 10:59:21 +02001595 struct page *pages[PIPE_BUFFERS];
1596 struct partial_page partial[PIPE_BUFFERS];
1597 struct splice_pipe_desc spd = {
1598 .pages = pages,
1599 .partial = partial,
1600 .flags = flags,
1601 .ops = &user_page_pipe_buf_ops,
Jens Axboebbdfc2f2007-11-06 23:29:47 -08001602 .spd_release = spd_release_page,
Jens Axboe912d35f2006-04-26 10:59:21 +02001603 };
1604
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001605 pipe = pipe_info(file->f_path.dentry->d_inode);
Jens Axboeddac0d32006-11-04 12:49:32 +01001606 if (!pipe)
Jens Axboe912d35f2006-04-26 10:59:21 +02001607 return -EBADF;
Jens Axboe912d35f2006-04-26 10:59:21 +02001608
Jens Axboe7afa6fd2006-05-01 20:02:33 +02001609 spd.nr_pages = get_iovec_page_array(iov, nr_segs, pages, partial,
1610 flags & SPLICE_F_GIFT);
Jens Axboe912d35f2006-04-26 10:59:21 +02001611 if (spd.nr_pages <= 0)
1612 return spd.nr_pages;
1613
Jens Axboe00522fb2006-04-26 14:39:29 +02001614 return splice_to_pipe(pipe, &spd);
Jens Axboe912d35f2006-04-26 10:59:21 +02001615}
1616
Jens Axboe6a14b902007-06-14 13:08:55 +02001617/*
1618 * Note that vmsplice only really supports true splicing _from_ user memory
1619 * to a pipe, not the other way around. Splicing from user memory is a simple
1620 * operation that can be supported without any funky alignment restrictions
1621 * or nasty vm tricks. We simply map in the user memory and fill them into
1622 * a pipe. The reverse isn't quite as easy, though. There are two possible
1623 * solutions for that:
1624 *
1625 * - memcpy() the data internally, at which point we might as well just
1626 * do a regular read() on the buffer anyway.
1627 * - Lots of nasty vm tricks, that are neither fast nor flexible (it
1628 * has restriction limitations on both ends of the pipe).
1629 *
1630 * Currently we punt and implement it as a normal copy, see pipe_to_user().
1631 *
1632 */
Heiko Carstens836f92a2009-01-14 14:14:33 +01001633SYSCALL_DEFINE4(vmsplice, int, fd, const struct iovec __user *, iov,
1634 unsigned long, nr_segs, unsigned int, flags)
Jens Axboe912d35f2006-04-26 10:59:21 +02001635{
1636 struct file *file;
1637 long error;
1638 int fput;
1639
Jens Axboe6a14b902007-06-14 13:08:55 +02001640 if (unlikely(nr_segs > UIO_MAXIOV))
1641 return -EINVAL;
1642 else if (unlikely(!nr_segs))
1643 return 0;
1644
Jens Axboe912d35f2006-04-26 10:59:21 +02001645 error = -EBADF;
1646 file = fget_light(fd, &fput);
1647 if (file) {
1648 if (file->f_mode & FMODE_WRITE)
Jens Axboe6a14b902007-06-14 13:08:55 +02001649 error = vmsplice_to_pipe(file, iov, nr_segs, flags);
1650 else if (file->f_mode & FMODE_READ)
1651 error = vmsplice_to_user(file, iov, nr_segs, flags);
Jens Axboe912d35f2006-04-26 10:59:21 +02001652
1653 fput_light(file, fput);
1654 }
1655
1656 return error;
1657}
1658
Heiko Carstens836f92a2009-01-14 14:14:33 +01001659SYSCALL_DEFINE6(splice, int, fd_in, loff_t __user *, off_in,
1660 int, fd_out, loff_t __user *, off_out,
1661 size_t, len, unsigned int, flags)
Jens Axboe5274f052006-03-30 15:15:30 +02001662{
1663 long error;
1664 struct file *in, *out;
1665 int fput_in, fput_out;
1666
1667 if (unlikely(!len))
1668 return 0;
1669
1670 error = -EBADF;
Ingo Molnar529565d2006-04-10 15:18:58 +02001671 in = fget_light(fd_in, &fput_in);
Jens Axboe5274f052006-03-30 15:15:30 +02001672 if (in) {
1673 if (in->f_mode & FMODE_READ) {
Ingo Molnar529565d2006-04-10 15:18:58 +02001674 out = fget_light(fd_out, &fput_out);
Jens Axboe5274f052006-03-30 15:15:30 +02001675 if (out) {
1676 if (out->f_mode & FMODE_WRITE)
Ingo Molnar529565d2006-04-10 15:18:58 +02001677 error = do_splice(in, off_in,
1678 out, off_out,
1679 len, flags);
Jens Axboe5274f052006-03-30 15:15:30 +02001680 fput_light(out, fput_out);
1681 }
1682 }
1683
1684 fput_light(in, fput_in);
1685 }
1686
1687 return error;
1688}
Jens Axboe70524492006-04-11 15:51:17 +02001689
1690/*
Jens Axboeaadd06e2006-07-10 11:00:01 +02001691 * Make sure there's data to read. Wait for input if we can, otherwise
1692 * return an appropriate error.
1693 */
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001694static int ipipe_prep(struct pipe_inode_info *pipe, unsigned int flags)
Jens Axboeaadd06e2006-07-10 11:00:01 +02001695{
1696 int ret;
1697
1698 /*
1699 * Check ->nrbufs without the inode lock first. This function
1700 * is speculative anyways, so missing one is ok.
1701 */
1702 if (pipe->nrbufs)
1703 return 0;
1704
1705 ret = 0;
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001706 pipe_lock(pipe);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001707
1708 while (!pipe->nrbufs) {
1709 if (signal_pending(current)) {
1710 ret = -ERESTARTSYS;
1711 break;
1712 }
1713 if (!pipe->writers)
1714 break;
1715 if (!pipe->waiting_writers) {
1716 if (flags & SPLICE_F_NONBLOCK) {
1717 ret = -EAGAIN;
1718 break;
1719 }
1720 }
1721 pipe_wait(pipe);
1722 }
1723
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001724 pipe_unlock(pipe);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001725 return ret;
1726}
1727
1728/*
1729 * Make sure there's writeable room. Wait for room if we can, otherwise
1730 * return an appropriate error.
1731 */
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001732static int opipe_prep(struct pipe_inode_info *pipe, unsigned int flags)
Jens Axboeaadd06e2006-07-10 11:00:01 +02001733{
1734 int ret;
1735
1736 /*
1737 * Check ->nrbufs without the inode lock first. This function
1738 * is speculative anyways, so missing one is ok.
1739 */
1740 if (pipe->nrbufs < PIPE_BUFFERS)
1741 return 0;
1742
1743 ret = 0;
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001744 pipe_lock(pipe);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001745
1746 while (pipe->nrbufs >= PIPE_BUFFERS) {
1747 if (!pipe->readers) {
1748 send_sig(SIGPIPE, current, 0);
1749 ret = -EPIPE;
1750 break;
1751 }
1752 if (flags & SPLICE_F_NONBLOCK) {
1753 ret = -EAGAIN;
1754 break;
1755 }
1756 if (signal_pending(current)) {
1757 ret = -ERESTARTSYS;
1758 break;
1759 }
1760 pipe->waiting_writers++;
1761 pipe_wait(pipe);
1762 pipe->waiting_writers--;
1763 }
1764
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001765 pipe_unlock(pipe);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001766 return ret;
1767}
1768
1769/*
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001770 * Splice contents of ipipe to opipe.
1771 */
1772static int splice_pipe_to_pipe(struct pipe_inode_info *ipipe,
1773 struct pipe_inode_info *opipe,
1774 size_t len, unsigned int flags)
1775{
1776 struct pipe_buffer *ibuf, *obuf;
1777 int ret = 0, nbuf;
1778 bool input_wakeup = false;
1779
1780
1781retry:
1782 ret = ipipe_prep(ipipe, flags);
1783 if (ret)
1784 return ret;
1785
1786 ret = opipe_prep(opipe, flags);
1787 if (ret)
1788 return ret;
1789
1790 /*
1791 * Potential ABBA deadlock, work around it by ordering lock
1792 * grabbing by pipe info address. Otherwise two different processes
1793 * could deadlock (one doing tee from A -> B, the other from B -> A).
1794 */
1795 pipe_double_lock(ipipe, opipe);
1796
1797 do {
1798 if (!opipe->readers) {
1799 send_sig(SIGPIPE, current, 0);
1800 if (!ret)
1801 ret = -EPIPE;
1802 break;
1803 }
1804
1805 if (!ipipe->nrbufs && !ipipe->writers)
1806 break;
1807
1808 /*
1809 * Cannot make any progress, because either the input
1810 * pipe is empty or the output pipe is full.
1811 */
1812 if (!ipipe->nrbufs || opipe->nrbufs >= PIPE_BUFFERS) {
1813 /* Already processed some buffers, break */
1814 if (ret)
1815 break;
1816
1817 if (flags & SPLICE_F_NONBLOCK) {
1818 ret = -EAGAIN;
1819 break;
1820 }
1821
1822 /*
1823 * We raced with another reader/writer and haven't
1824 * managed to process any buffers. A zero return
1825 * value means EOF, so retry instead.
1826 */
1827 pipe_unlock(ipipe);
1828 pipe_unlock(opipe);
1829 goto retry;
1830 }
1831
1832 ibuf = ipipe->bufs + ipipe->curbuf;
1833 nbuf = (opipe->curbuf + opipe->nrbufs) % PIPE_BUFFERS;
1834 obuf = opipe->bufs + nbuf;
1835
1836 if (len >= ibuf->len) {
1837 /*
1838 * Simply move the whole buffer from ipipe to opipe
1839 */
1840 *obuf = *ibuf;
1841 ibuf->ops = NULL;
1842 opipe->nrbufs++;
1843 ipipe->curbuf = (ipipe->curbuf + 1) % PIPE_BUFFERS;
1844 ipipe->nrbufs--;
1845 input_wakeup = true;
1846 } else {
1847 /*
1848 * Get a reference to this pipe buffer,
1849 * so we can copy the contents over.
1850 */
1851 ibuf->ops->get(ipipe, ibuf);
1852 *obuf = *ibuf;
1853
1854 /*
1855 * Don't inherit the gift flag, we need to
1856 * prevent multiple steals of this page.
1857 */
1858 obuf->flags &= ~PIPE_BUF_FLAG_GIFT;
1859
1860 obuf->len = len;
1861 opipe->nrbufs++;
1862 ibuf->offset += obuf->len;
1863 ibuf->len -= obuf->len;
1864 }
1865 ret += obuf->len;
1866 len -= obuf->len;
1867 } while (len);
1868
1869 pipe_unlock(ipipe);
1870 pipe_unlock(opipe);
1871
1872 /*
1873 * If we put data in the output pipe, wakeup any potential readers.
1874 */
1875 if (ret > 0) {
1876 smp_mb();
1877 if (waitqueue_active(&opipe->wait))
1878 wake_up_interruptible(&opipe->wait);
1879 kill_fasync(&opipe->fasync_readers, SIGIO, POLL_IN);
1880 }
1881 if (input_wakeup)
1882 wakeup_pipe_writers(ipipe);
1883
1884 return ret;
1885}
1886
1887/*
Jens Axboe70524492006-04-11 15:51:17 +02001888 * Link contents of ipipe to opipe.
1889 */
1890static int link_pipe(struct pipe_inode_info *ipipe,
1891 struct pipe_inode_info *opipe,
1892 size_t len, unsigned int flags)
1893{
1894 struct pipe_buffer *ibuf, *obuf;
Jens Axboeaadd06e2006-07-10 11:00:01 +02001895 int ret = 0, i = 0, nbuf;
Jens Axboe70524492006-04-11 15:51:17 +02001896
1897 /*
1898 * Potential ABBA deadlock, work around it by ordering lock
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001899 * grabbing by pipe info address. Otherwise two different processes
Jens Axboe70524492006-04-11 15:51:17 +02001900 * could deadlock (one doing tee from A -> B, the other from B -> A).
1901 */
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001902 pipe_double_lock(ipipe, opipe);
Jens Axboe70524492006-04-11 15:51:17 +02001903
Jens Axboeaadd06e2006-07-10 11:00:01 +02001904 do {
Jens Axboe70524492006-04-11 15:51:17 +02001905 if (!opipe->readers) {
1906 send_sig(SIGPIPE, current, 0);
1907 if (!ret)
1908 ret = -EPIPE;
1909 break;
1910 }
Jens Axboe70524492006-04-11 15:51:17 +02001911
1912 /*
Jens Axboeaadd06e2006-07-10 11:00:01 +02001913 * If we have iterated all input buffers or ran out of
1914 * output room, break.
Jens Axboe70524492006-04-11 15:51:17 +02001915 */
Jens Axboeaadd06e2006-07-10 11:00:01 +02001916 if (i >= ipipe->nrbufs || opipe->nrbufs >= PIPE_BUFFERS)
Jens Axboe70524492006-04-11 15:51:17 +02001917 break;
Jens Axboeaadd06e2006-07-10 11:00:01 +02001918
1919 ibuf = ipipe->bufs + ((ipipe->curbuf + i) & (PIPE_BUFFERS - 1));
1920 nbuf = (opipe->curbuf + opipe->nrbufs) & (PIPE_BUFFERS - 1);
1921
Jens Axboe2a27250e2006-04-19 15:56:40 +02001922 /*
Jens Axboeaadd06e2006-07-10 11:00:01 +02001923 * Get a reference to this pipe buffer,
1924 * so we can copy the contents over.
Jens Axboe2a27250e2006-04-19 15:56:40 +02001925 */
Jens Axboeaadd06e2006-07-10 11:00:01 +02001926 ibuf->ops->get(ipipe, ibuf);
Jens Axboe70524492006-04-11 15:51:17 +02001927
Jens Axboeaadd06e2006-07-10 11:00:01 +02001928 obuf = opipe->bufs + nbuf;
1929 *obuf = *ibuf;
Jens Axboe70524492006-04-11 15:51:17 +02001930
Jens Axboeaadd06e2006-07-10 11:00:01 +02001931 /*
1932 * Don't inherit the gift flag, we need to
1933 * prevent multiple steals of this page.
1934 */
1935 obuf->flags &= ~PIPE_BUF_FLAG_GIFT;
1936
1937 if (obuf->len > len)
1938 obuf->len = len;
1939
1940 opipe->nrbufs++;
1941 ret += obuf->len;
1942 len -= obuf->len;
1943 i++;
1944 } while (len);
Jens Axboe70524492006-04-11 15:51:17 +02001945
Jens Axboe02cf01a2008-02-20 10:34:51 +01001946 /*
1947 * return EAGAIN if we have the potential of some data in the
1948 * future, otherwise just return 0
1949 */
1950 if (!ret && ipipe->waiting_writers && (flags & SPLICE_F_NONBLOCK))
1951 ret = -EAGAIN;
1952
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001953 pipe_unlock(ipipe);
1954 pipe_unlock(opipe);
Jens Axboe70524492006-04-11 15:51:17 +02001955
Jens Axboeaadd06e2006-07-10 11:00:01 +02001956 /*
1957 * If we put data in the output pipe, wakeup any potential readers.
1958 */
1959 if (ret > 0) {
Jens Axboe70524492006-04-11 15:51:17 +02001960 smp_mb();
1961 if (waitqueue_active(&opipe->wait))
1962 wake_up_interruptible(&opipe->wait);
1963 kill_fasync(&opipe->fasync_readers, SIGIO, POLL_IN);
1964 }
1965
1966 return ret;
1967}
1968
1969/*
1970 * This is a tee(1) implementation that works on pipes. It doesn't copy
1971 * any data, it simply references the 'in' pages on the 'out' pipe.
1972 * The 'flags' used are the SPLICE_F_* variants, currently the only
1973 * applicable one is SPLICE_F_NONBLOCK.
1974 */
1975static long do_tee(struct file *in, struct file *out, size_t len,
1976 unsigned int flags)
1977{
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001978 struct pipe_inode_info *ipipe = pipe_info(in->f_path.dentry->d_inode);
1979 struct pipe_inode_info *opipe = pipe_info(out->f_path.dentry->d_inode);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001980 int ret = -EINVAL;
Jens Axboe70524492006-04-11 15:51:17 +02001981
1982 /*
Jens Axboeaadd06e2006-07-10 11:00:01 +02001983 * Duplicate the contents of ipipe to opipe without actually
1984 * copying the data.
Jens Axboe70524492006-04-11 15:51:17 +02001985 */
Jens Axboeaadd06e2006-07-10 11:00:01 +02001986 if (ipipe && opipe && ipipe != opipe) {
1987 /*
1988 * Keep going, unless we encounter an error. The ipipe/opipe
1989 * ordering doesn't really matter.
1990 */
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001991 ret = ipipe_prep(ipipe, flags);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001992 if (!ret) {
Miklos Szeredi7c77f0b2009-05-07 15:37:35 +02001993 ret = opipe_prep(opipe, flags);
Jens Axboe02cf01a2008-02-20 10:34:51 +01001994 if (!ret)
Jens Axboeaadd06e2006-07-10 11:00:01 +02001995 ret = link_pipe(ipipe, opipe, len, flags);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001996 }
1997 }
Jens Axboe70524492006-04-11 15:51:17 +02001998
Jens Axboeaadd06e2006-07-10 11:00:01 +02001999 return ret;
Jens Axboe70524492006-04-11 15:51:17 +02002000}
2001
Heiko Carstens836f92a2009-01-14 14:14:33 +01002002SYSCALL_DEFINE4(tee, int, fdin, int, fdout, size_t, len, unsigned int, flags)
Jens Axboe70524492006-04-11 15:51:17 +02002003{
2004 struct file *in;
2005 int error, fput_in;
2006
2007 if (unlikely(!len))
2008 return 0;
2009
2010 error = -EBADF;
2011 in = fget_light(fdin, &fput_in);
2012 if (in) {
2013 if (in->f_mode & FMODE_READ) {
2014 int fput_out;
2015 struct file *out = fget_light(fdout, &fput_out);
2016
2017 if (out) {
2018 if (out->f_mode & FMODE_WRITE)
2019 error = do_tee(in, out, len, flags);
2020 fput_light(out, fput_out);
2021 }
2022 }
2023 fput_light(in, fput_in);
2024 }
2025
2026 return error;
2027}