blob: 666953d59a35c77670514b916aea7515fe612951 [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}
510
Jens Axboe059a8f32006-04-02 23:06:05 +0200511EXPORT_SYMBOL(generic_file_splice_read);
512
Jens Axboe5274f052006-03-30 15:15:30 +0200513/*
Jens Axboe4f6f0bd2006-04-02 23:04:46 +0200514 * Send 'sd->len' bytes to socket from 'sd->file' at position 'sd->pos'
Jens Axboe016b6612006-04-25 15:42:00 +0200515 * using sendpage(). Return the number of bytes sent.
Jens Axboe5274f052006-03-30 15:15:30 +0200516 */
Jens Axboe76ad4d12006-05-03 10:41:33 +0200517static int pipe_to_sendpage(struct pipe_inode_info *pipe,
Jens Axboe5274f052006-03-30 15:15:30 +0200518 struct pipe_buffer *buf, struct splice_desc *sd)
519{
Jens Axboe6a14b902007-06-14 13:08:55 +0200520 struct file *file = sd->u.file;
Jens Axboe5274f052006-03-30 15:15:30 +0200521 loff_t pos = sd->pos;
Jens Axboef84d7512006-05-01 19:59:03 +0200522 int ret, more;
Jens Axboe5274f052006-03-30 15:15:30 +0200523
Jens Axboecac36bb2007-06-14 13:10:48 +0200524 ret = buf->ops->confirm(pipe, buf);
Jens Axboef84d7512006-05-01 19:59:03 +0200525 if (!ret) {
526 more = (sd->flags & SPLICE_F_MORE) || sd->len < sd->total_len;
Jens Axboe5274f052006-03-30 15:15:30 +0200527
Jens Axboef84d7512006-05-01 19:59:03 +0200528 ret = file->f_op->sendpage(file, buf->page, buf->offset,
529 sd->len, &pos, more);
530 }
Jens Axboe5274f052006-03-30 15:15:30 +0200531
Jens Axboe016b6612006-04-25 15:42:00 +0200532 return ret;
Jens Axboe5274f052006-03-30 15:15:30 +0200533}
534
535/*
536 * This is a little more tricky than the file -> pipe splicing. There are
537 * basically three cases:
538 *
539 * - Destination page already exists in the address space and there
540 * are users of it. For that case we have no other option that
541 * copying the data. Tough luck.
542 * - Destination page already exists in the address space, but there
543 * are no users of it. Make sure it's uptodate, then drop it. Fall
544 * through to last case.
545 * - Destination page does not exist, we can add the pipe page to
546 * the page cache and avoid the copy.
547 *
Jens Axboe83f91352006-04-02 23:05:09 +0200548 * If asked to move pages to the output file (SPLICE_F_MOVE is set in
549 * sd->flags), we attempt to migrate pages from the pipe to the output
550 * file address space page cache. This is possible if no one else has
551 * the pipe page referenced outside of the pipe and page cache. If
552 * SPLICE_F_MOVE isn't set, or we cannot move the page, we simply create
553 * a new page in the output file page cache and fill/dirty that.
Jens Axboe5274f052006-03-30 15:15:30 +0200554 */
Miklos Szeredi328eaab2009-04-14 19:48:39 +0200555int pipe_to_file(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
556 struct splice_desc *sd)
Jens Axboe5274f052006-03-30 15:15:30 +0200557{
Jens Axboe6a14b902007-06-14 13:08:55 +0200558 struct file *file = sd->u.file;
Jens Axboe5274f052006-03-30 15:15:30 +0200559 struct address_space *mapping = file->f_mapping;
Jens Axboe016b6612006-04-25 15:42:00 +0200560 unsigned int offset, this_len;
Jens Axboe5274f052006-03-30 15:15:30 +0200561 struct page *page;
Nick Pigginafddba42007-10-16 01:25:01 -0700562 void *fsdata;
Jens Axboe3e7ee3e2006-04-02 23:11:04 +0200563 int ret;
Jens Axboe5274f052006-03-30 15:15:30 +0200564
565 /*
Jens Axboe49d0b212006-04-10 09:04:41 +0200566 * make sure the data in this buffer is uptodate
Jens Axboe5274f052006-03-30 15:15:30 +0200567 */
Jens Axboecac36bb2007-06-14 13:10:48 +0200568 ret = buf->ops->confirm(pipe, buf);
Jens Axboef84d7512006-05-01 19:59:03 +0200569 if (unlikely(ret))
570 return ret;
Jens Axboe5274f052006-03-30 15:15:30 +0200571
Jens Axboe5274f052006-03-30 15:15:30 +0200572 offset = sd->pos & ~PAGE_CACHE_MASK;
573
Jens Axboe016b6612006-04-25 15:42:00 +0200574 this_len = sd->len;
575 if (this_len + offset > PAGE_CACHE_SIZE)
576 this_len = PAGE_CACHE_SIZE - offset;
577
Nick Pigginafddba42007-10-16 01:25:01 -0700578 ret = pagecache_write_begin(file, mapping, sd->pos, this_len,
579 AOP_FLAG_UNINTERRUPTIBLE, &page, &fsdata);
580 if (unlikely(ret))
581 goto out;
Jens Axboe5274f052006-03-30 15:15:30 +0200582
Jens Axboe0568b402006-05-01 19:50:48 +0200583 if (buf->page != page) {
Jens Axboef84d7512006-05-01 19:59:03 +0200584 /*
585 * Careful, ->map() uses KM_USER0!
586 */
Jens Axboe76ad4d12006-05-03 10:41:33 +0200587 char *src = buf->ops->map(pipe, buf, 1);
Jens Axboef84d7512006-05-01 19:59:03 +0200588 char *dst = kmap_atomic(page, KM_USER1);
Jens Axboe5abc97a2006-03-30 15:16:46 +0200589
Jens Axboe016b6612006-04-25 15:42:00 +0200590 memcpy(dst + offset, src + buf->offset, this_len);
Jens Axboe5abc97a2006-03-30 15:16:46 +0200591 flush_dcache_page(page);
Jens Axboef84d7512006-05-01 19:59:03 +0200592 kunmap_atomic(dst, KM_USER1);
Jens Axboe76ad4d12006-05-03 10:41:33 +0200593 buf->ops->unmap(pipe, buf, src);
Jens Axboe5abc97a2006-03-30 15:16:46 +0200594 }
Nick Pigginafddba42007-10-16 01:25:01 -0700595 ret = pagecache_write_end(file, mapping, sd->pos, this_len, this_len,
596 page, fsdata);
Jens Axboe5274f052006-03-30 15:15:30 +0200597out:
Jens Axboe5274f052006-03-30 15:15:30 +0200598 return ret;
599}
Miklos Szeredi328eaab2009-04-14 19:48:39 +0200600EXPORT_SYMBOL(pipe_to_file);
Jens Axboe5274f052006-03-30 15:15:30 +0200601
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200602static void wakeup_pipe_writers(struct pipe_inode_info *pipe)
603{
604 smp_mb();
605 if (waitqueue_active(&pipe->wait))
606 wake_up_interruptible(&pipe->wait);
607 kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
608}
609
610/**
611 * splice_from_pipe_feed - feed available data from a pipe to a file
612 * @pipe: pipe to splice from
613 * @sd: information to @actor
614 * @actor: handler that splices the data
615 *
616 * Description:
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200617 * This function loops over the pipe and calls @actor to do the
618 * actual moving of a single struct pipe_buffer to the desired
619 * destination. It returns when there's no more buffers left in
620 * the pipe or if the requested number of bytes (@sd->total_len)
621 * have been copied. It returns a positive number (one) if the
622 * pipe needs to be filled with more data, zero if the required
623 * number of bytes have been copied and -errno on error.
624 *
625 * This, together with splice_from_pipe_{begin,end,next}, may be
626 * used to implement the functionality of __splice_from_pipe() when
627 * locking is required around copying the pipe buffers to the
628 * destination.
629 */
630int splice_from_pipe_feed(struct pipe_inode_info *pipe, struct splice_desc *sd,
631 splice_actor *actor)
632{
633 int ret;
634
635 while (pipe->nrbufs) {
636 struct pipe_buffer *buf = pipe->bufs + pipe->curbuf;
637 const struct pipe_buf_operations *ops = buf->ops;
638
639 sd->len = buf->len;
640 if (sd->len > sd->total_len)
641 sd->len = sd->total_len;
642
643 ret = actor(pipe, buf, sd);
644 if (ret <= 0) {
645 if (ret == -ENODATA)
646 ret = 0;
647 return ret;
648 }
649 buf->offset += ret;
650 buf->len -= ret;
651
652 sd->num_spliced += ret;
653 sd->len -= ret;
654 sd->pos += ret;
655 sd->total_len -= ret;
656
657 if (!buf->len) {
658 buf->ops = NULL;
659 ops->release(pipe, buf);
660 pipe->curbuf = (pipe->curbuf + 1) & (PIPE_BUFFERS - 1);
661 pipe->nrbufs--;
662 if (pipe->inode)
663 sd->need_wakeup = true;
664 }
665
666 if (!sd->total_len)
667 return 0;
668 }
669
670 return 1;
671}
672EXPORT_SYMBOL(splice_from_pipe_feed);
673
674/**
675 * splice_from_pipe_next - wait for some data to splice from
676 * @pipe: pipe to splice from
677 * @sd: information about the splice operation
678 *
679 * Description:
680 * This function will wait for some data and return a positive
681 * value (one) if pipe buffers are available. It will return zero
682 * or -errno if no more data needs to be spliced.
683 */
684int splice_from_pipe_next(struct pipe_inode_info *pipe, struct splice_desc *sd)
685{
686 while (!pipe->nrbufs) {
687 if (!pipe->writers)
688 return 0;
689
690 if (!pipe->waiting_writers && sd->num_spliced)
691 return 0;
692
693 if (sd->flags & SPLICE_F_NONBLOCK)
694 return -EAGAIN;
695
696 if (signal_pending(current))
697 return -ERESTARTSYS;
698
699 if (sd->need_wakeup) {
700 wakeup_pipe_writers(pipe);
701 sd->need_wakeup = false;
702 }
703
704 pipe_wait(pipe);
705 }
706
707 return 1;
708}
709EXPORT_SYMBOL(splice_from_pipe_next);
710
711/**
712 * splice_from_pipe_begin - start splicing from pipe
Randy Dunlapb80901b2009-04-16 19:09:55 -0700713 * @sd: information about the splice operation
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200714 *
715 * Description:
716 * This function should be called before a loop containing
717 * splice_from_pipe_next() and splice_from_pipe_feed() to
718 * initialize the necessary fields of @sd.
719 */
720void splice_from_pipe_begin(struct splice_desc *sd)
721{
722 sd->num_spliced = 0;
723 sd->need_wakeup = false;
724}
725EXPORT_SYMBOL(splice_from_pipe_begin);
726
727/**
728 * splice_from_pipe_end - finish splicing from pipe
729 * @pipe: pipe to splice from
730 * @sd: information about the splice operation
731 *
732 * Description:
733 * This function will wake up pipe writers if necessary. It should
734 * be called after a loop containing splice_from_pipe_next() and
735 * splice_from_pipe_feed().
736 */
737void splice_from_pipe_end(struct pipe_inode_info *pipe, struct splice_desc *sd)
738{
739 if (sd->need_wakeup)
740 wakeup_pipe_writers(pipe);
741}
742EXPORT_SYMBOL(splice_from_pipe_end);
743
Jens Axboe932cc6d2007-06-21 13:10:21 +0200744/**
745 * __splice_from_pipe - splice data from a pipe to given actor
746 * @pipe: pipe to splice from
747 * @sd: information to @actor
748 * @actor: handler that splices the data
749 *
750 * Description:
751 * This function does little more than loop over the pipe and call
752 * @actor to do the actual moving of a single struct pipe_buffer to
753 * the desired destination. See pipe_to_file, pipe_to_sendpage, or
754 * pipe_to_user.
755 *
Jens Axboe83f91352006-04-02 23:05:09 +0200756 */
Jens Axboec66ab6f2007-06-12 21:17:17 +0200757ssize_t __splice_from_pipe(struct pipe_inode_info *pipe, struct splice_desc *sd,
758 splice_actor *actor)
Jens Axboe5274f052006-03-30 15:15:30 +0200759{
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200760 int ret;
Jens Axboe5274f052006-03-30 15:15:30 +0200761
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200762 splice_from_pipe_begin(sd);
763 do {
764 ret = splice_from_pipe_next(pipe, sd);
765 if (ret > 0)
766 ret = splice_from_pipe_feed(pipe, sd, actor);
767 } while (ret > 0);
768 splice_from_pipe_end(pipe, sd);
Jens Axboe5274f052006-03-30 15:15:30 +0200769
Miklos Szeredib3c2d2d2009-04-14 19:48:36 +0200770 return sd->num_spliced ? sd->num_spliced : ret;
Jens Axboe5274f052006-03-30 15:15:30 +0200771}
Mark Fasheh40bee442007-03-21 13:11:02 +0100772EXPORT_SYMBOL(__splice_from_pipe);
Jens Axboe5274f052006-03-30 15:15:30 +0200773
Jens Axboe932cc6d2007-06-21 13:10:21 +0200774/**
775 * splice_from_pipe - splice data from a pipe to a file
776 * @pipe: pipe to splice from
777 * @out: file to splice to
778 * @ppos: position in @out
779 * @len: how many bytes to splice
780 * @flags: splice modifier flags
781 * @actor: handler that splices the data
782 *
783 * Description:
Miklos Szeredi29339702009-04-14 19:48:37 +0200784 * See __splice_from_pipe. This function locks the pipe inode,
Jens Axboe932cc6d2007-06-21 13:10:21 +0200785 * otherwise it's identical to __splice_from_pipe().
786 *
787 */
Mark Fasheh6da61802006-10-17 18:43:07 +0200788ssize_t splice_from_pipe(struct pipe_inode_info *pipe, struct file *out,
789 loff_t *ppos, size_t len, unsigned int flags,
790 splice_actor *actor)
791{
792 ssize_t ret;
Jens Axboec66ab6f2007-06-12 21:17:17 +0200793 struct splice_desc sd = {
794 .total_len = len,
795 .flags = flags,
796 .pos = *ppos,
Jens Axboe6a14b902007-06-14 13:08:55 +0200797 .u.file = out,
Jens Axboec66ab6f2007-06-12 21:17:17 +0200798 };
Mark Fasheh6da61802006-10-17 18:43:07 +0200799
Miklos Szeredi61e0d472009-04-14 19:48:41 +0200800 pipe_lock(pipe);
Jens Axboec66ab6f2007-06-12 21:17:17 +0200801 ret = __splice_from_pipe(pipe, &sd, actor);
Miklos Szeredi61e0d472009-04-14 19:48:41 +0200802 pipe_unlock(pipe);
Mark Fasheh6da61802006-10-17 18:43:07 +0200803
804 return ret;
805}
806
807/**
Jens Axboe83f91352006-04-02 23:05:09 +0200808 * generic_file_splice_write - splice data from a pipe to a file
Ingo Molnar3a326a22006-04-10 15:18:35 +0200809 * @pipe: pipe info
Jens Axboe83f91352006-04-02 23:05:09 +0200810 * @out: file to write to
Jens Axboe932cc6d2007-06-21 13:10:21 +0200811 * @ppos: position in @out
Jens Axboe83f91352006-04-02 23:05:09 +0200812 * @len: number of bytes to splice
813 * @flags: splice modifier flags
814 *
Jens Axboe932cc6d2007-06-21 13:10:21 +0200815 * Description:
816 * Will either move or copy pages (determined by @flags options) from
817 * the given pipe inode to the given file.
Jens Axboe83f91352006-04-02 23:05:09 +0200818 *
819 */
Ingo Molnar3a326a22006-04-10 15:18:35 +0200820ssize_t
821generic_file_splice_write(struct pipe_inode_info *pipe, struct file *out,
Jens Axboecbb7e572006-04-11 14:57:50 +0200822 loff_t *ppos, size_t len, unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +0200823{
Jens Axboe4f6f0bd2006-04-02 23:04:46 +0200824 struct address_space *mapping = out->f_mapping;
Jens Axboe8c34e2d2006-10-17 19:43:22 +0200825 struct inode *inode = mapping->host;
Miklos Szeredi7f3d4ee2008-05-07 09:22:39 +0200826 struct splice_desc sd = {
827 .total_len = len,
828 .flags = flags,
829 .pos = *ppos,
830 .u.file = out,
831 };
Ingo Molnar3a326a22006-04-10 15:18:35 +0200832 ssize_t ret;
Jens Axboe8c34e2d2006-10-17 19:43:22 +0200833
Miklos Szeredi61e0d472009-04-14 19:48:41 +0200834 pipe_lock(pipe);
Miklos Szeredieb443e52009-04-14 19:48:38 +0200835
836 splice_from_pipe_begin(&sd);
837 do {
838 ret = splice_from_pipe_next(pipe, &sd);
839 if (ret <= 0)
840 break;
841
842 mutex_lock_nested(&inode->i_mutex, I_MUTEX_CHILD);
843 ret = file_remove_suid(out);
844 if (!ret)
845 ret = splice_from_pipe_feed(pipe, &sd, pipe_to_file);
846 mutex_unlock(&inode->i_mutex);
847 } while (ret > 0);
848 splice_from_pipe_end(pipe, &sd);
849
Miklos Szeredi61e0d472009-04-14 19:48:41 +0200850 pipe_unlock(pipe);
Miklos Szeredieb443e52009-04-14 19:48:38 +0200851
852 if (sd.num_spliced)
853 ret = sd.num_spliced;
854
Jens Axboea4514eb2006-04-19 15:57:05 +0200855 if (ret > 0) {
Jens Axboe17ee4f42007-06-15 13:10:37 +0200856 unsigned long nr_pages;
857
Jens Axboea4514eb2006-04-19 15:57:05 +0200858 *ppos += ret;
Jens Axboe17ee4f42007-06-15 13:10:37 +0200859 nr_pages = (ret + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
Jens Axboe4f6f0bd2006-04-02 23:04:46 +0200860
Jens Axboea4514eb2006-04-19 15:57:05 +0200861 /*
862 * If file or inode is SYNC and we actually wrote some data,
863 * sync it.
864 */
865 if (unlikely((out->f_flags & O_SYNC) || IS_SYNC(inode))) {
Miklos Szeredi7f3d4ee2008-05-07 09:22:39 +0200866 int err;
867
Jens Axboea4514eb2006-04-19 15:57:05 +0200868 mutex_lock(&inode->i_mutex);
869 err = generic_osync_inode(inode, mapping,
870 OSYNC_METADATA|OSYNC_DATA);
871 mutex_unlock(&inode->i_mutex);
872
873 if (err)
874 ret = err;
875 }
Jens Axboe17ee4f42007-06-15 13:10:37 +0200876 balance_dirty_pages_ratelimited_nr(mapping, nr_pages);
Jens Axboe4f6f0bd2006-04-02 23:04:46 +0200877 }
878
879 return ret;
Jens Axboe5274f052006-03-30 15:15:30 +0200880}
881
Jens Axboe059a8f32006-04-02 23:06:05 +0200882EXPORT_SYMBOL(generic_file_splice_write);
883
Jens Axboe83f91352006-04-02 23:05:09 +0200884/**
885 * generic_splice_sendpage - splice data from a pipe to a socket
Jens Axboe932cc6d2007-06-21 13:10:21 +0200886 * @pipe: pipe to splice from
Jens Axboe83f91352006-04-02 23:05:09 +0200887 * @out: socket to write to
Jens Axboe932cc6d2007-06-21 13:10:21 +0200888 * @ppos: position in @out
Jens Axboe83f91352006-04-02 23:05:09 +0200889 * @len: number of bytes to splice
890 * @flags: splice modifier flags
891 *
Jens Axboe932cc6d2007-06-21 13:10:21 +0200892 * Description:
893 * Will send @len bytes from the pipe to a network socket. No data copying
894 * is involved.
Jens Axboe83f91352006-04-02 23:05:09 +0200895 *
896 */
Ingo Molnar3a326a22006-04-10 15:18:35 +0200897ssize_t generic_splice_sendpage(struct pipe_inode_info *pipe, struct file *out,
Jens Axboecbb7e572006-04-11 14:57:50 +0200898 loff_t *ppos, size_t len, unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +0200899{
Jens Axboe00522fb2006-04-26 14:39:29 +0200900 return splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_sendpage);
Jens Axboe5274f052006-03-30 15:15:30 +0200901}
902
Jens Axboe059a8f32006-04-02 23:06:05 +0200903EXPORT_SYMBOL(generic_splice_sendpage);
Jeff Garzika0f06782006-03-30 23:06:13 -0500904
Jens Axboe83f91352006-04-02 23:05:09 +0200905/*
906 * Attempt to initiate a splice from pipe to file.
907 */
Ingo Molnar3a326a22006-04-10 15:18:35 +0200908static long do_splice_from(struct pipe_inode_info *pipe, struct file *out,
Jens Axboecbb7e572006-04-11 14:57:50 +0200909 loff_t *ppos, size_t len, unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +0200910{
Jens Axboe5274f052006-03-30 15:15:30 +0200911 int ret;
912
Jens Axboe49570e92006-04-11 13:56:09 +0200913 if (unlikely(!out->f_op || !out->f_op->splice_write))
Jens Axboe5274f052006-03-30 15:15:30 +0200914 return -EINVAL;
915
Jens Axboe49570e92006-04-11 13:56:09 +0200916 if (unlikely(!(out->f_mode & FMODE_WRITE)))
Jens Axboe5274f052006-03-30 15:15:30 +0200917 return -EBADF;
918
Linus Torvaldsefc968d2008-10-09 14:04:54 -0700919 if (unlikely(out->f_flags & O_APPEND))
920 return -EINVAL;
921
Jens Axboecbb7e572006-04-11 14:57:50 +0200922 ret = rw_verify_area(WRITE, out, ppos, len);
Jens Axboe5274f052006-03-30 15:15:30 +0200923 if (unlikely(ret < 0))
924 return ret;
925
Jens Axboecbb7e572006-04-11 14:57:50 +0200926 return out->f_op->splice_write(pipe, out, ppos, len, flags);
Jens Axboe5274f052006-03-30 15:15:30 +0200927}
928
Jens Axboe83f91352006-04-02 23:05:09 +0200929/*
930 * Attempt to initiate a splice from a file to a pipe.
931 */
Jens Axboecbb7e572006-04-11 14:57:50 +0200932static long do_splice_to(struct file *in, loff_t *ppos,
933 struct pipe_inode_info *pipe, size_t len,
934 unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +0200935{
Jens Axboe5274f052006-03-30 15:15:30 +0200936 int ret;
937
Jens Axboe49570e92006-04-11 13:56:09 +0200938 if (unlikely(!in->f_op || !in->f_op->splice_read))
Jens Axboe5274f052006-03-30 15:15:30 +0200939 return -EINVAL;
940
Jens Axboe49570e92006-04-11 13:56:09 +0200941 if (unlikely(!(in->f_mode & FMODE_READ)))
Jens Axboe5274f052006-03-30 15:15:30 +0200942 return -EBADF;
943
Jens Axboecbb7e572006-04-11 14:57:50 +0200944 ret = rw_verify_area(READ, in, ppos, len);
Jens Axboe5274f052006-03-30 15:15:30 +0200945 if (unlikely(ret < 0))
946 return ret;
947
Jens Axboecbb7e572006-04-11 14:57:50 +0200948 return in->f_op->splice_read(in, ppos, pipe, len, flags);
Jens Axboe5274f052006-03-30 15:15:30 +0200949}
950
Jens Axboe932cc6d2007-06-21 13:10:21 +0200951/**
952 * splice_direct_to_actor - splices data directly between two non-pipes
953 * @in: file to splice from
954 * @sd: actor information on where to splice to
955 * @actor: handles the data splicing
956 *
957 * Description:
958 * This is a special case helper to splice directly between two
959 * points, without requiring an explicit pipe. Internally an allocated
Randy Dunlap79685b82007-07-27 08:08:51 +0200960 * pipe is cached in the process, and reused during the lifetime of
Jens Axboe932cc6d2007-06-21 13:10:21 +0200961 * that process.
962 *
Jens Axboec66ab6f2007-06-12 21:17:17 +0200963 */
964ssize_t splice_direct_to_actor(struct file *in, struct splice_desc *sd,
965 splice_direct_actor *actor)
Jens Axboeb92ce552006-04-11 13:52:07 +0200966{
967 struct pipe_inode_info *pipe;
968 long ret, bytes;
969 umode_t i_mode;
Jens Axboec66ab6f2007-06-12 21:17:17 +0200970 size_t len;
971 int i, flags;
Jens Axboeb92ce552006-04-11 13:52:07 +0200972
973 /*
974 * We require the input being a regular file, as we don't want to
975 * randomly drop data for eg socket -> socket splicing. Use the
976 * piped splicing for that!
977 */
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800978 i_mode = in->f_path.dentry->d_inode->i_mode;
Jens Axboeb92ce552006-04-11 13:52:07 +0200979 if (unlikely(!S_ISREG(i_mode) && !S_ISBLK(i_mode)))
980 return -EINVAL;
981
982 /*
983 * neither in nor out is a pipe, setup an internal pipe attached to
984 * 'out' and transfer the wanted data from 'in' to 'out' through that
985 */
986 pipe = current->splice_pipe;
Jens Axboe49570e92006-04-11 13:56:09 +0200987 if (unlikely(!pipe)) {
Jens Axboeb92ce552006-04-11 13:52:07 +0200988 pipe = alloc_pipe_info(NULL);
989 if (!pipe)
990 return -ENOMEM;
991
992 /*
993 * We don't have an immediate reader, but we'll read the stuff
Jens Axboe00522fb2006-04-26 14:39:29 +0200994 * out of the pipe right after the splice_to_pipe(). So set
Jens Axboeb92ce552006-04-11 13:52:07 +0200995 * PIPE_READERS appropriately.
996 */
997 pipe->readers = 1;
998
999 current->splice_pipe = pipe;
1000 }
1001
1002 /*
Ingo Molnar73d62d82006-04-11 13:57:21 +02001003 * Do the splice.
Jens Axboeb92ce552006-04-11 13:52:07 +02001004 */
1005 ret = 0;
1006 bytes = 0;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001007 len = sd->total_len;
1008 flags = sd->flags;
1009
1010 /*
1011 * Don't block on output, we have to drain the direct pipe.
1012 */
1013 sd->flags &= ~SPLICE_F_NONBLOCK;
Jens Axboeb92ce552006-04-11 13:52:07 +02001014
1015 while (len) {
Jens Axboe51a92c02007-07-13 14:11:43 +02001016 size_t read_len;
Tom Zanussia82c53a2008-05-09 13:28:36 +02001017 loff_t pos = sd->pos, prev_pos = pos;
Jens Axboeb92ce552006-04-11 13:52:07 +02001018
Jens Axboebcd4f3a2007-07-16 14:41:49 +02001019 ret = do_splice_to(in, &pos, pipe, len, flags);
Jens Axboe51a92c02007-07-13 14:11:43 +02001020 if (unlikely(ret <= 0))
Jens Axboeb92ce552006-04-11 13:52:07 +02001021 goto out_release;
1022
1023 read_len = ret;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001024 sd->total_len = read_len;
Jens Axboeb92ce552006-04-11 13:52:07 +02001025
1026 /*
1027 * NOTE: nonblocking mode only applies to the input. We
1028 * must not do the output in nonblocking mode as then we
1029 * could get stuck data in the internal pipe:
1030 */
Jens Axboec66ab6f2007-06-12 21:17:17 +02001031 ret = actor(pipe, sd);
Tom Zanussia82c53a2008-05-09 13:28:36 +02001032 if (unlikely(ret <= 0)) {
1033 sd->pos = prev_pos;
Jens Axboeb92ce552006-04-11 13:52:07 +02001034 goto out_release;
Tom Zanussia82c53a2008-05-09 13:28:36 +02001035 }
Jens Axboeb92ce552006-04-11 13:52:07 +02001036
1037 bytes += ret;
1038 len -= ret;
Jens Axboebcd4f3a2007-07-16 14:41:49 +02001039 sd->pos = pos;
Jens Axboeb92ce552006-04-11 13:52:07 +02001040
Tom Zanussia82c53a2008-05-09 13:28:36 +02001041 if (ret < read_len) {
1042 sd->pos = prev_pos + ret;
Jens Axboe51a92c02007-07-13 14:11:43 +02001043 goto out_release;
Tom Zanussia82c53a2008-05-09 13:28:36 +02001044 }
Jens Axboeb92ce552006-04-11 13:52:07 +02001045 }
1046
Jens Axboe9e971982008-01-29 21:05:57 +01001047done:
Jens Axboeb92ce552006-04-11 13:52:07 +02001048 pipe->nrbufs = pipe->curbuf = 0;
Jens Axboe80848702008-01-30 12:24:48 +01001049 file_accessed(in);
Jens Axboeb92ce552006-04-11 13:52:07 +02001050 return bytes;
1051
1052out_release:
1053 /*
1054 * If we did an incomplete transfer we must release
1055 * the pipe buffers in question:
1056 */
1057 for (i = 0; i < PIPE_BUFFERS; i++) {
1058 struct pipe_buffer *buf = pipe->bufs + i;
1059
1060 if (buf->ops) {
1061 buf->ops->release(pipe, buf);
1062 buf->ops = NULL;
1063 }
1064 }
Jens Axboeb92ce552006-04-11 13:52:07 +02001065
Jens Axboe9e971982008-01-29 21:05:57 +01001066 if (!bytes)
1067 bytes = ret;
Jens Axboeb92ce552006-04-11 13:52:07 +02001068
Jens Axboe9e971982008-01-29 21:05:57 +01001069 goto done;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001070}
1071EXPORT_SYMBOL(splice_direct_to_actor);
1072
1073static int direct_splice_actor(struct pipe_inode_info *pipe,
1074 struct splice_desc *sd)
1075{
Jens Axboe6a14b902007-06-14 13:08:55 +02001076 struct file *file = sd->u.file;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001077
1078 return do_splice_from(pipe, file, &sd->pos, sd->total_len, sd->flags);
1079}
1080
Jens Axboe932cc6d2007-06-21 13:10:21 +02001081/**
1082 * do_splice_direct - splices data directly between two files
1083 * @in: file to splice from
1084 * @ppos: input file offset
1085 * @out: file to splice to
1086 * @len: number of bytes to splice
1087 * @flags: splice modifier flags
1088 *
1089 * Description:
1090 * For use by do_sendfile(). splice can easily emulate sendfile, but
1091 * doing it in the application would incur an extra system call
1092 * (splice in + splice out, as compared to just sendfile()). So this helper
1093 * can splice directly through a process-private pipe.
1094 *
1095 */
Jens Axboec66ab6f2007-06-12 21:17:17 +02001096long do_splice_direct(struct file *in, loff_t *ppos, struct file *out,
1097 size_t len, unsigned int flags)
1098{
1099 struct splice_desc sd = {
1100 .len = len,
1101 .total_len = len,
1102 .flags = flags,
1103 .pos = *ppos,
Jens Axboe6a14b902007-06-14 13:08:55 +02001104 .u.file = out,
Jens Axboec66ab6f2007-06-12 21:17:17 +02001105 };
Jens Axboe51a92c02007-07-13 14:11:43 +02001106 long ret;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001107
1108 ret = splice_direct_to_actor(in, &sd, direct_splice_actor);
Jens Axboe51a92c02007-07-13 14:11:43 +02001109 if (ret > 0)
Tom Zanussia82c53a2008-05-09 13:28:36 +02001110 *ppos = sd.pos;
Jens Axboe51a92c02007-07-13 14:11:43 +02001111
Jens Axboec66ab6f2007-06-12 21:17:17 +02001112 return ret;
Jens Axboeb92ce552006-04-11 13:52:07 +02001113}
1114
Jens Axboe83f91352006-04-02 23:05:09 +02001115/*
Jens Axboeddac0d32006-11-04 12:49:32 +01001116 * After the inode slimming patch, i_pipe/i_bdev/i_cdev share the same
1117 * location, so checking ->i_pipe is not enough to verify that this is a
1118 * pipe.
1119 */
1120static inline struct pipe_inode_info *pipe_info(struct inode *inode)
1121{
1122 if (S_ISFIFO(inode->i_mode))
1123 return inode->i_pipe;
1124
1125 return NULL;
1126}
1127
1128/*
Jens Axboe83f91352006-04-02 23:05:09 +02001129 * Determine where to splice to/from.
1130 */
Ingo Molnar529565d2006-04-10 15:18:58 +02001131static long do_splice(struct file *in, loff_t __user *off_in,
1132 struct file *out, loff_t __user *off_out,
1133 size_t len, unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +02001134{
Ingo Molnar3a326a22006-04-10 15:18:35 +02001135 struct pipe_inode_info *pipe;
Jens Axboecbb7e572006-04-11 14:57:50 +02001136 loff_t offset, *off;
Jens Axboea4514eb2006-04-19 15:57:05 +02001137 long ret;
Jens Axboe5274f052006-03-30 15:15:30 +02001138
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001139 pipe = pipe_info(in->f_path.dentry->d_inode);
Ingo Molnar529565d2006-04-10 15:18:58 +02001140 if (pipe) {
1141 if (off_in)
1142 return -ESPIPE;
Jens Axboeb92ce552006-04-11 13:52:07 +02001143 if (off_out) {
1144 if (out->f_op->llseek == no_llseek)
1145 return -EINVAL;
Jens Axboecbb7e572006-04-11 14:57:50 +02001146 if (copy_from_user(&offset, off_out, sizeof(loff_t)))
Jens Axboeb92ce552006-04-11 13:52:07 +02001147 return -EFAULT;
Jens Axboecbb7e572006-04-11 14:57:50 +02001148 off = &offset;
1149 } else
1150 off = &out->f_pos;
Ingo Molnar529565d2006-04-10 15:18:58 +02001151
Jens Axboea4514eb2006-04-19 15:57:05 +02001152 ret = do_splice_from(pipe, out, off, len, flags);
1153
1154 if (off_out && copy_to_user(off_out, off, sizeof(loff_t)))
1155 ret = -EFAULT;
1156
1157 return ret;
Ingo Molnar529565d2006-04-10 15:18:58 +02001158 }
Jens Axboe5274f052006-03-30 15:15:30 +02001159
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001160 pipe = pipe_info(out->f_path.dentry->d_inode);
Ingo Molnar529565d2006-04-10 15:18:58 +02001161 if (pipe) {
1162 if (off_out)
1163 return -ESPIPE;
Jens Axboeb92ce552006-04-11 13:52:07 +02001164 if (off_in) {
1165 if (in->f_op->llseek == no_llseek)
1166 return -EINVAL;
Jens Axboecbb7e572006-04-11 14:57:50 +02001167 if (copy_from_user(&offset, off_in, sizeof(loff_t)))
Jens Axboeb92ce552006-04-11 13:52:07 +02001168 return -EFAULT;
Jens Axboecbb7e572006-04-11 14:57:50 +02001169 off = &offset;
1170 } else
1171 off = &in->f_pos;
Ingo Molnar529565d2006-04-10 15:18:58 +02001172
Jens Axboea4514eb2006-04-19 15:57:05 +02001173 ret = do_splice_to(in, off, pipe, len, flags);
1174
1175 if (off_in && copy_to_user(off_in, off, sizeof(loff_t)))
1176 ret = -EFAULT;
1177
1178 return ret;
Ingo Molnar529565d2006-04-10 15:18:58 +02001179 }
Jens Axboe5274f052006-03-30 15:15:30 +02001180
1181 return -EINVAL;
1182}
1183
Jens Axboe912d35f2006-04-26 10:59:21 +02001184/*
1185 * Map an iov into an array of pages and offset/length tupples. With the
1186 * partial_page structure, we can map several non-contiguous ranges into
1187 * our ones pages[] map instead of splitting that operation into pieces.
1188 * Could easily be exported as a generic helper for other users, in which
1189 * case one would probably want to add a 'max_nr_pages' parameter as well.
1190 */
1191static int get_iovec_page_array(const struct iovec __user *iov,
1192 unsigned int nr_vecs, struct page **pages,
Jens Axboe7afa6fd2006-05-01 20:02:33 +02001193 struct partial_page *partial, int aligned)
Jens Axboe912d35f2006-04-26 10:59:21 +02001194{
1195 int buffers = 0, error = 0;
1196
Jens Axboe912d35f2006-04-26 10:59:21 +02001197 while (nr_vecs) {
1198 unsigned long off, npages;
Linus Torvalds75723952007-10-01 13:17:28 -07001199 struct iovec entry;
Jens Axboe912d35f2006-04-26 10:59:21 +02001200 void __user *base;
1201 size_t len;
1202 int i;
1203
Linus Torvalds75723952007-10-01 13:17:28 -07001204 error = -EFAULT;
Nick Pigginbc40d732008-07-25 19:45:26 -07001205 if (copy_from_user(&entry, iov, sizeof(entry)))
Jens Axboe912d35f2006-04-26 10:59:21 +02001206 break;
Linus Torvalds75723952007-10-01 13:17:28 -07001207
1208 base = entry.iov_base;
1209 len = entry.iov_len;
Jens Axboe912d35f2006-04-26 10:59:21 +02001210
1211 /*
1212 * Sanity check this iovec. 0 read succeeds.
1213 */
Linus Torvalds75723952007-10-01 13:17:28 -07001214 error = 0;
Jens Axboe912d35f2006-04-26 10:59:21 +02001215 if (unlikely(!len))
1216 break;
1217 error = -EFAULT;
Bastian Blank712a30e2008-02-10 16:47:57 +02001218 if (!access_ok(VERIFY_READ, base, len))
Jens Axboe912d35f2006-04-26 10:59:21 +02001219 break;
1220
1221 /*
1222 * Get this base offset and number of pages, then map
1223 * in the user pages.
1224 */
1225 off = (unsigned long) base & ~PAGE_MASK;
Jens Axboe7afa6fd2006-05-01 20:02:33 +02001226
1227 /*
1228 * If asked for alignment, the offset must be zero and the
1229 * length a multiple of the PAGE_SIZE.
1230 */
1231 error = -EINVAL;
1232 if (aligned && (off || len & ~PAGE_MASK))
1233 break;
1234
Jens Axboe912d35f2006-04-26 10:59:21 +02001235 npages = (off + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1236 if (npages > PIPE_BUFFERS - buffers)
1237 npages = PIPE_BUFFERS - buffers;
1238
Nick Pigginbc40d732008-07-25 19:45:26 -07001239 error = get_user_pages_fast((unsigned long)base, npages,
1240 0, &pages[buffers]);
Jens Axboe912d35f2006-04-26 10:59:21 +02001241
1242 if (unlikely(error <= 0))
1243 break;
1244
1245 /*
1246 * Fill this contiguous range into the partial page map.
1247 */
1248 for (i = 0; i < error; i++) {
Jens Axboe75914892006-05-02 12:57:18 +02001249 const int plen = min_t(size_t, len, PAGE_SIZE - off);
Jens Axboe912d35f2006-04-26 10:59:21 +02001250
1251 partial[buffers].offset = off;
1252 partial[buffers].len = plen;
1253
1254 off = 0;
1255 len -= plen;
1256 buffers++;
1257 }
1258
1259 /*
1260 * We didn't complete this iov, stop here since it probably
1261 * means we have to move some of this into a pipe to
1262 * be able to continue.
1263 */
1264 if (len)
1265 break;
1266
1267 /*
1268 * Don't continue if we mapped fewer pages than we asked for,
1269 * or if we mapped the max number of pages that we have
1270 * room for.
1271 */
1272 if (error < npages || buffers == PIPE_BUFFERS)
1273 break;
1274
1275 nr_vecs--;
1276 iov++;
1277 }
1278
Jens Axboe912d35f2006-04-26 10:59:21 +02001279 if (buffers)
1280 return buffers;
1281
1282 return error;
1283}
1284
Jens Axboe6a14b902007-06-14 13:08:55 +02001285static int pipe_to_user(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
1286 struct splice_desc *sd)
1287{
1288 char *src;
1289 int ret;
1290
Jens Axboecac36bb2007-06-14 13:10:48 +02001291 ret = buf->ops->confirm(pipe, buf);
Jens Axboe6a14b902007-06-14 13:08:55 +02001292 if (unlikely(ret))
1293 return ret;
1294
1295 /*
1296 * See if we can use the atomic maps, by prefaulting in the
1297 * pages and doing an atomic copy
1298 */
1299 if (!fault_in_pages_writeable(sd->u.userptr, sd->len)) {
1300 src = buf->ops->map(pipe, buf, 1);
1301 ret = __copy_to_user_inatomic(sd->u.userptr, src + buf->offset,
1302 sd->len);
1303 buf->ops->unmap(pipe, buf, src);
1304 if (!ret) {
1305 ret = sd->len;
1306 goto out;
1307 }
1308 }
1309
1310 /*
1311 * No dice, use slow non-atomic map and copy
1312 */
1313 src = buf->ops->map(pipe, buf, 0);
1314
1315 ret = sd->len;
1316 if (copy_to_user(sd->u.userptr, src + buf->offset, sd->len))
1317 ret = -EFAULT;
1318
Jens Axboe6866bef2007-10-16 10:01:29 +02001319 buf->ops->unmap(pipe, buf, src);
Jens Axboe6a14b902007-06-14 13:08:55 +02001320out:
1321 if (ret > 0)
1322 sd->u.userptr += ret;
Jens Axboe6a14b902007-06-14 13:08:55 +02001323 return ret;
1324}
1325
1326/*
1327 * For lack of a better implementation, implement vmsplice() to userspace
1328 * as a simple copy of the pipes pages to the user iov.
1329 */
1330static long vmsplice_to_user(struct file *file, const struct iovec __user *iov,
1331 unsigned long nr_segs, unsigned int flags)
1332{
1333 struct pipe_inode_info *pipe;
1334 struct splice_desc sd;
1335 ssize_t size;
1336 int error;
1337 long ret;
1338
1339 pipe = pipe_info(file->f_path.dentry->d_inode);
1340 if (!pipe)
1341 return -EBADF;
1342
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001343 pipe_lock(pipe);
Jens Axboe6a14b902007-06-14 13:08:55 +02001344
1345 error = ret = 0;
1346 while (nr_segs) {
1347 void __user *base;
1348 size_t len;
1349
1350 /*
1351 * Get user address base and length for this iovec.
1352 */
1353 error = get_user(base, &iov->iov_base);
1354 if (unlikely(error))
1355 break;
1356 error = get_user(len, &iov->iov_len);
1357 if (unlikely(error))
1358 break;
1359
1360 /*
1361 * Sanity check this iovec. 0 read succeeds.
1362 */
1363 if (unlikely(!len))
1364 break;
1365 if (unlikely(!base)) {
1366 error = -EFAULT;
1367 break;
1368 }
1369
Jens Axboe88119302008-02-08 08:49:14 -08001370 if (unlikely(!access_ok(VERIFY_WRITE, base, len))) {
1371 error = -EFAULT;
1372 break;
1373 }
1374
Jens Axboe6a14b902007-06-14 13:08:55 +02001375 sd.len = 0;
1376 sd.total_len = len;
1377 sd.flags = flags;
1378 sd.u.userptr = base;
1379 sd.pos = 0;
1380
1381 size = __splice_from_pipe(pipe, &sd, pipe_to_user);
1382 if (size < 0) {
1383 if (!ret)
1384 ret = size;
1385
1386 break;
1387 }
1388
1389 ret += size;
1390
1391 if (size < len)
1392 break;
1393
1394 nr_segs--;
1395 iov++;
1396 }
1397
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001398 pipe_unlock(pipe);
Jens Axboe6a14b902007-06-14 13:08:55 +02001399
1400 if (!ret)
1401 ret = error;
1402
1403 return ret;
1404}
1405
Jens Axboe912d35f2006-04-26 10:59:21 +02001406/*
1407 * vmsplice splices a user address range into a pipe. It can be thought of
1408 * as splice-from-memory, where the regular splice is splice-from-file (or
1409 * to file). In both cases the output is a pipe, naturally.
Jens Axboe912d35f2006-04-26 10:59:21 +02001410 */
Jens Axboe6a14b902007-06-14 13:08:55 +02001411static long vmsplice_to_pipe(struct file *file, const struct iovec __user *iov,
1412 unsigned long nr_segs, unsigned int flags)
Jens Axboe912d35f2006-04-26 10:59:21 +02001413{
Jens Axboeddac0d32006-11-04 12:49:32 +01001414 struct pipe_inode_info *pipe;
Jens Axboe912d35f2006-04-26 10:59:21 +02001415 struct page *pages[PIPE_BUFFERS];
1416 struct partial_page partial[PIPE_BUFFERS];
1417 struct splice_pipe_desc spd = {
1418 .pages = pages,
1419 .partial = partial,
1420 .flags = flags,
1421 .ops = &user_page_pipe_buf_ops,
Jens Axboebbdfc2f2007-11-06 23:29:47 -08001422 .spd_release = spd_release_page,
Jens Axboe912d35f2006-04-26 10:59:21 +02001423 };
1424
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001425 pipe = pipe_info(file->f_path.dentry->d_inode);
Jens Axboeddac0d32006-11-04 12:49:32 +01001426 if (!pipe)
Jens Axboe912d35f2006-04-26 10:59:21 +02001427 return -EBADF;
Jens Axboe912d35f2006-04-26 10:59:21 +02001428
Jens Axboe7afa6fd2006-05-01 20:02:33 +02001429 spd.nr_pages = get_iovec_page_array(iov, nr_segs, pages, partial,
1430 flags & SPLICE_F_GIFT);
Jens Axboe912d35f2006-04-26 10:59:21 +02001431 if (spd.nr_pages <= 0)
1432 return spd.nr_pages;
1433
Jens Axboe00522fb2006-04-26 14:39:29 +02001434 return splice_to_pipe(pipe, &spd);
Jens Axboe912d35f2006-04-26 10:59:21 +02001435}
1436
Jens Axboe6a14b902007-06-14 13:08:55 +02001437/*
1438 * Note that vmsplice only really supports true splicing _from_ user memory
1439 * to a pipe, not the other way around. Splicing from user memory is a simple
1440 * operation that can be supported without any funky alignment restrictions
1441 * or nasty vm tricks. We simply map in the user memory and fill them into
1442 * a pipe. The reverse isn't quite as easy, though. There are two possible
1443 * solutions for that:
1444 *
1445 * - memcpy() the data internally, at which point we might as well just
1446 * do a regular read() on the buffer anyway.
1447 * - Lots of nasty vm tricks, that are neither fast nor flexible (it
1448 * has restriction limitations on both ends of the pipe).
1449 *
1450 * Currently we punt and implement it as a normal copy, see pipe_to_user().
1451 *
1452 */
Heiko Carstens836f92a2009-01-14 14:14:33 +01001453SYSCALL_DEFINE4(vmsplice, int, fd, const struct iovec __user *, iov,
1454 unsigned long, nr_segs, unsigned int, flags)
Jens Axboe912d35f2006-04-26 10:59:21 +02001455{
1456 struct file *file;
1457 long error;
1458 int fput;
1459
Jens Axboe6a14b902007-06-14 13:08:55 +02001460 if (unlikely(nr_segs > UIO_MAXIOV))
1461 return -EINVAL;
1462 else if (unlikely(!nr_segs))
1463 return 0;
1464
Jens Axboe912d35f2006-04-26 10:59:21 +02001465 error = -EBADF;
1466 file = fget_light(fd, &fput);
1467 if (file) {
1468 if (file->f_mode & FMODE_WRITE)
Jens Axboe6a14b902007-06-14 13:08:55 +02001469 error = vmsplice_to_pipe(file, iov, nr_segs, flags);
1470 else if (file->f_mode & FMODE_READ)
1471 error = vmsplice_to_user(file, iov, nr_segs, flags);
Jens Axboe912d35f2006-04-26 10:59:21 +02001472
1473 fput_light(file, fput);
1474 }
1475
1476 return error;
1477}
1478
Heiko Carstens836f92a2009-01-14 14:14:33 +01001479SYSCALL_DEFINE6(splice, int, fd_in, loff_t __user *, off_in,
1480 int, fd_out, loff_t __user *, off_out,
1481 size_t, len, unsigned int, flags)
Jens Axboe5274f052006-03-30 15:15:30 +02001482{
1483 long error;
1484 struct file *in, *out;
1485 int fput_in, fput_out;
1486
1487 if (unlikely(!len))
1488 return 0;
1489
1490 error = -EBADF;
Ingo Molnar529565d2006-04-10 15:18:58 +02001491 in = fget_light(fd_in, &fput_in);
Jens Axboe5274f052006-03-30 15:15:30 +02001492 if (in) {
1493 if (in->f_mode & FMODE_READ) {
Ingo Molnar529565d2006-04-10 15:18:58 +02001494 out = fget_light(fd_out, &fput_out);
Jens Axboe5274f052006-03-30 15:15:30 +02001495 if (out) {
1496 if (out->f_mode & FMODE_WRITE)
Ingo Molnar529565d2006-04-10 15:18:58 +02001497 error = do_splice(in, off_in,
1498 out, off_out,
1499 len, flags);
Jens Axboe5274f052006-03-30 15:15:30 +02001500 fput_light(out, fput_out);
1501 }
1502 }
1503
1504 fput_light(in, fput_in);
1505 }
1506
1507 return error;
1508}
Jens Axboe70524492006-04-11 15:51:17 +02001509
1510/*
Jens Axboeaadd06e2006-07-10 11:00:01 +02001511 * Make sure there's data to read. Wait for input if we can, otherwise
1512 * return an appropriate error.
1513 */
1514static int link_ipipe_prep(struct pipe_inode_info *pipe, unsigned int flags)
1515{
1516 int ret;
1517
1518 /*
1519 * Check ->nrbufs without the inode lock first. This function
1520 * is speculative anyways, so missing one is ok.
1521 */
1522 if (pipe->nrbufs)
1523 return 0;
1524
1525 ret = 0;
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001526 pipe_lock(pipe);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001527
1528 while (!pipe->nrbufs) {
1529 if (signal_pending(current)) {
1530 ret = -ERESTARTSYS;
1531 break;
1532 }
1533 if (!pipe->writers)
1534 break;
1535 if (!pipe->waiting_writers) {
1536 if (flags & SPLICE_F_NONBLOCK) {
1537 ret = -EAGAIN;
1538 break;
1539 }
1540 }
1541 pipe_wait(pipe);
1542 }
1543
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001544 pipe_unlock(pipe);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001545 return ret;
1546}
1547
1548/*
1549 * Make sure there's writeable room. Wait for room if we can, otherwise
1550 * return an appropriate error.
1551 */
1552static int link_opipe_prep(struct pipe_inode_info *pipe, unsigned int flags)
1553{
1554 int ret;
1555
1556 /*
1557 * Check ->nrbufs without the inode lock first. This function
1558 * is speculative anyways, so missing one is ok.
1559 */
1560 if (pipe->nrbufs < PIPE_BUFFERS)
1561 return 0;
1562
1563 ret = 0;
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001564 pipe_lock(pipe);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001565
1566 while (pipe->nrbufs >= PIPE_BUFFERS) {
1567 if (!pipe->readers) {
1568 send_sig(SIGPIPE, current, 0);
1569 ret = -EPIPE;
1570 break;
1571 }
1572 if (flags & SPLICE_F_NONBLOCK) {
1573 ret = -EAGAIN;
1574 break;
1575 }
1576 if (signal_pending(current)) {
1577 ret = -ERESTARTSYS;
1578 break;
1579 }
1580 pipe->waiting_writers++;
1581 pipe_wait(pipe);
1582 pipe->waiting_writers--;
1583 }
1584
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001585 pipe_unlock(pipe);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001586 return ret;
1587}
1588
1589/*
Jens Axboe70524492006-04-11 15:51:17 +02001590 * Link contents of ipipe to opipe.
1591 */
1592static int link_pipe(struct pipe_inode_info *ipipe,
1593 struct pipe_inode_info *opipe,
1594 size_t len, unsigned int flags)
1595{
1596 struct pipe_buffer *ibuf, *obuf;
Jens Axboeaadd06e2006-07-10 11:00:01 +02001597 int ret = 0, i = 0, nbuf;
Jens Axboe70524492006-04-11 15:51:17 +02001598
1599 /*
1600 * Potential ABBA deadlock, work around it by ordering lock
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001601 * grabbing by pipe info address. Otherwise two different processes
Jens Axboe70524492006-04-11 15:51:17 +02001602 * could deadlock (one doing tee from A -> B, the other from B -> A).
1603 */
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001604 pipe_double_lock(ipipe, opipe);
Jens Axboe70524492006-04-11 15:51:17 +02001605
Jens Axboeaadd06e2006-07-10 11:00:01 +02001606 do {
Jens Axboe70524492006-04-11 15:51:17 +02001607 if (!opipe->readers) {
1608 send_sig(SIGPIPE, current, 0);
1609 if (!ret)
1610 ret = -EPIPE;
1611 break;
1612 }
Jens Axboe70524492006-04-11 15:51:17 +02001613
1614 /*
Jens Axboeaadd06e2006-07-10 11:00:01 +02001615 * If we have iterated all input buffers or ran out of
1616 * output room, break.
Jens Axboe70524492006-04-11 15:51:17 +02001617 */
Jens Axboeaadd06e2006-07-10 11:00:01 +02001618 if (i >= ipipe->nrbufs || opipe->nrbufs >= PIPE_BUFFERS)
Jens Axboe70524492006-04-11 15:51:17 +02001619 break;
Jens Axboeaadd06e2006-07-10 11:00:01 +02001620
1621 ibuf = ipipe->bufs + ((ipipe->curbuf + i) & (PIPE_BUFFERS - 1));
1622 nbuf = (opipe->curbuf + opipe->nrbufs) & (PIPE_BUFFERS - 1);
1623
Jens Axboe2a272502006-04-19 15:56:40 +02001624 /*
Jens Axboeaadd06e2006-07-10 11:00:01 +02001625 * Get a reference to this pipe buffer,
1626 * so we can copy the contents over.
Jens Axboe2a272502006-04-19 15:56:40 +02001627 */
Jens Axboeaadd06e2006-07-10 11:00:01 +02001628 ibuf->ops->get(ipipe, ibuf);
Jens Axboe70524492006-04-11 15:51:17 +02001629
Jens Axboeaadd06e2006-07-10 11:00:01 +02001630 obuf = opipe->bufs + nbuf;
1631 *obuf = *ibuf;
Jens Axboe70524492006-04-11 15:51:17 +02001632
Jens Axboeaadd06e2006-07-10 11:00:01 +02001633 /*
1634 * Don't inherit the gift flag, we need to
1635 * prevent multiple steals of this page.
1636 */
1637 obuf->flags &= ~PIPE_BUF_FLAG_GIFT;
1638
1639 if (obuf->len > len)
1640 obuf->len = len;
1641
1642 opipe->nrbufs++;
1643 ret += obuf->len;
1644 len -= obuf->len;
1645 i++;
1646 } while (len);
Jens Axboe70524492006-04-11 15:51:17 +02001647
Jens Axboe02cf01a2008-02-20 10:34:51 +01001648 /*
1649 * return EAGAIN if we have the potential of some data in the
1650 * future, otherwise just return 0
1651 */
1652 if (!ret && ipipe->waiting_writers && (flags & SPLICE_F_NONBLOCK))
1653 ret = -EAGAIN;
1654
Miklos Szeredi61e0d472009-04-14 19:48:41 +02001655 pipe_unlock(ipipe);
1656 pipe_unlock(opipe);
Jens Axboe70524492006-04-11 15:51:17 +02001657
Jens Axboeaadd06e2006-07-10 11:00:01 +02001658 /*
1659 * If we put data in the output pipe, wakeup any potential readers.
1660 */
1661 if (ret > 0) {
Jens Axboe70524492006-04-11 15:51:17 +02001662 smp_mb();
1663 if (waitqueue_active(&opipe->wait))
1664 wake_up_interruptible(&opipe->wait);
1665 kill_fasync(&opipe->fasync_readers, SIGIO, POLL_IN);
1666 }
1667
1668 return ret;
1669}
1670
1671/*
1672 * This is a tee(1) implementation that works on pipes. It doesn't copy
1673 * any data, it simply references the 'in' pages on the 'out' pipe.
1674 * The 'flags' used are the SPLICE_F_* variants, currently the only
1675 * applicable one is SPLICE_F_NONBLOCK.
1676 */
1677static long do_tee(struct file *in, struct file *out, size_t len,
1678 unsigned int flags)
1679{
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001680 struct pipe_inode_info *ipipe = pipe_info(in->f_path.dentry->d_inode);
1681 struct pipe_inode_info *opipe = pipe_info(out->f_path.dentry->d_inode);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001682 int ret = -EINVAL;
Jens Axboe70524492006-04-11 15:51:17 +02001683
1684 /*
Jens Axboeaadd06e2006-07-10 11:00:01 +02001685 * Duplicate the contents of ipipe to opipe without actually
1686 * copying the data.
Jens Axboe70524492006-04-11 15:51:17 +02001687 */
Jens Axboeaadd06e2006-07-10 11:00:01 +02001688 if (ipipe && opipe && ipipe != opipe) {
1689 /*
1690 * Keep going, unless we encounter an error. The ipipe/opipe
1691 * ordering doesn't really matter.
1692 */
1693 ret = link_ipipe_prep(ipipe, flags);
1694 if (!ret) {
1695 ret = link_opipe_prep(opipe, flags);
Jens Axboe02cf01a2008-02-20 10:34:51 +01001696 if (!ret)
Jens Axboeaadd06e2006-07-10 11:00:01 +02001697 ret = link_pipe(ipipe, opipe, len, flags);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001698 }
1699 }
Jens Axboe70524492006-04-11 15:51:17 +02001700
Jens Axboeaadd06e2006-07-10 11:00:01 +02001701 return ret;
Jens Axboe70524492006-04-11 15:51:17 +02001702}
1703
Heiko Carstens836f92a2009-01-14 14:14:33 +01001704SYSCALL_DEFINE4(tee, int, fdin, int, fdout, size_t, len, unsigned int, flags)
Jens Axboe70524492006-04-11 15:51:17 +02001705{
1706 struct file *in;
1707 int error, fput_in;
1708
1709 if (unlikely(!len))
1710 return 0;
1711
1712 error = -EBADF;
1713 in = fget_light(fdin, &fput_in);
1714 if (in) {
1715 if (in->f_mode & FMODE_READ) {
1716 int fput_out;
1717 struct file *out = fget_light(fdout, &fput_out);
1718
1719 if (out) {
1720 if (out->f_mode & FMODE_WRITE)
1721 error = do_tee(in, out, len, flags);
1722 fput_light(out, fput_out);
1723 }
1724 }
1725 fput_light(in, fput_in);
1726 }
1727
1728 return error;
1729}