blob: 78150038b58422155b2a89e47e8a228967f0ca3d [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>
Jens Axboe5274f052006-03-30 15:15:30 +020024#include <linux/mm_inline.h>
Jens Axboe5abc97a2006-03-30 15:16:46 +020025#include <linux/swap.h>
Jens Axboe4f6f0bd2006-04-02 23:04:46 +020026#include <linux/writeback.h>
27#include <linux/buffer_head.h>
Jeff Garzika0f06782006-03-30 23:06:13 -050028#include <linux/module.h>
Jens Axboe4f6f0bd2006-04-02 23:04:46 +020029#include <linux/syscalls.h>
Jens Axboe912d35f2006-04-26 10:59:21 +020030#include <linux/uio.h>
James Morris29ce2052007-07-13 11:44:32 +020031#include <linux/security.h>
Jens Axboe5274f052006-03-30 15:15:30 +020032
Jens Axboe83f91352006-04-02 23:05:09 +020033/*
34 * Attempt to steal a page from a pipe buffer. This should perhaps go into
35 * a vm helper function, it's already simplified quite a bit by the
36 * addition of remove_mapping(). If success is returned, the caller may
37 * attempt to reuse this page for another destination.
38 */
Jens Axboe76ad4d12006-05-03 10:41:33 +020039static int page_cache_pipe_buf_steal(struct pipe_inode_info *pipe,
Jens Axboe5abc97a2006-03-30 15:16:46 +020040 struct pipe_buffer *buf)
41{
42 struct page *page = buf->page;
Jens Axboe9e94cd42006-06-20 15:01:12 +020043 struct address_space *mapping;
Jens Axboe5abc97a2006-03-30 15:16:46 +020044
Jens Axboe9e0267c2006-04-19 15:57:31 +020045 lock_page(page);
46
Jens Axboe9e94cd42006-06-20 15:01:12 +020047 mapping = page_mapping(page);
48 if (mapping) {
49 WARN_ON(!PageUptodate(page));
Jens Axboe5abc97a2006-03-30 15:16:46 +020050
Jens Axboe9e94cd42006-06-20 15:01:12 +020051 /*
52 * At least for ext2 with nobh option, we need to wait on
53 * writeback completing on this page, since we'll remove it
54 * from the pagecache. Otherwise truncate wont wait on the
55 * page, allowing the disk blocks to be reused by someone else
56 * before we actually wrote our data to them. fs corruption
57 * ensues.
58 */
59 wait_on_page_writeback(page);
Jens Axboead8d6f02006-04-02 23:10:32 +020060
Jens Axboe9e94cd42006-06-20 15:01:12 +020061 if (PagePrivate(page))
Nick Piggin2ae88142006-10-28 10:38:23 -070062 try_to_release_page(page, GFP_KERNEL);
Jens Axboe4f6f0bd2006-04-02 23:04:46 +020063
Jens Axboe9e94cd42006-06-20 15:01:12 +020064 /*
65 * If we succeeded in removing the mapping, set LRU flag
66 * and return good.
67 */
68 if (remove_mapping(mapping, page)) {
69 buf->flags |= PIPE_BUF_FLAG_LRU;
70 return 0;
71 }
Jens Axboe9e0267c2006-04-19 15:57:31 +020072 }
Jens Axboe5abc97a2006-03-30 15:16:46 +020073
Jens Axboe9e94cd42006-06-20 15:01:12 +020074 /*
75 * Raced with truncate or failed to remove page from current
76 * address space, unlock and return failure.
77 */
78 unlock_page(page);
79 return 1;
Jens Axboe5abc97a2006-03-30 15:16:46 +020080}
81
Jens Axboe76ad4d12006-05-03 10:41:33 +020082static void page_cache_pipe_buf_release(struct pipe_inode_info *pipe,
Jens Axboe5274f052006-03-30 15:15:30 +020083 struct pipe_buffer *buf)
84{
85 page_cache_release(buf->page);
Jens Axboe14328732006-05-03 10:35:26 +020086 buf->flags &= ~PIPE_BUF_FLAG_LRU;
Jens Axboe5274f052006-03-30 15:15:30 +020087}
88
Jens Axboe08457182007-06-12 20:51:32 +020089/*
90 * Check whether the contents of buf is OK to access. Since the content
91 * is a page cache page, IO may be in flight.
92 */
Jens Axboecac36bb02007-06-14 13:10:48 +020093static int page_cache_pipe_buf_confirm(struct pipe_inode_info *pipe,
94 struct pipe_buffer *buf)
Jens Axboe5274f052006-03-30 15:15:30 +020095{
96 struct page *page = buf->page;
Jens Axboe49d0b212006-04-10 09:04:41 +020097 int err;
Jens Axboe5274f052006-03-30 15:15:30 +020098
99 if (!PageUptodate(page)) {
Jens Axboe49d0b212006-04-10 09:04:41 +0200100 lock_page(page);
101
102 /*
103 * Page got truncated/unhashed. This will cause a 0-byte
Ingo Molnar73d62d82006-04-11 13:57:21 +0200104 * splice, if this is the first page.
Jens Axboe49d0b212006-04-10 09:04:41 +0200105 */
106 if (!page->mapping) {
107 err = -ENODATA;
108 goto error;
109 }
110
111 /*
Ingo Molnar73d62d82006-04-11 13:57:21 +0200112 * Uh oh, read-error from disk.
Jens Axboe49d0b212006-04-10 09:04:41 +0200113 */
114 if (!PageUptodate(page)) {
115 err = -EIO;
116 goto error;
117 }
118
119 /*
Jens Axboef84d7512006-05-01 19:59:03 +0200120 * Page is ok afterall, we are done.
Jens Axboe49d0b212006-04-10 09:04:41 +0200121 */
Jens Axboe5274f052006-03-30 15:15:30 +0200122 unlock_page(page);
Jens Axboe5274f052006-03-30 15:15:30 +0200123 }
124
Jens Axboef84d7512006-05-01 19:59:03 +0200125 return 0;
Jens Axboe49d0b212006-04-10 09:04:41 +0200126error:
127 unlock_page(page);
Jens Axboef84d7512006-05-01 19:59:03 +0200128 return err;
Jens Axboe70524492006-04-11 15:51:17 +0200129}
130
Eric Dumazetd4c3cca2006-12-13 00:34:04 -0800131static const struct pipe_buf_operations page_cache_pipe_buf_ops = {
Jens Axboe5274f052006-03-30 15:15:30 +0200132 .can_merge = 0,
Jens Axboef84d7512006-05-01 19:59:03 +0200133 .map = generic_pipe_buf_map,
134 .unmap = generic_pipe_buf_unmap,
Jens Axboecac36bb02007-06-14 13:10:48 +0200135 .confirm = page_cache_pipe_buf_confirm,
Jens Axboe5274f052006-03-30 15:15:30 +0200136 .release = page_cache_pipe_buf_release,
Jens Axboe5abc97a2006-03-30 15:16:46 +0200137 .steal = page_cache_pipe_buf_steal,
Jens Axboef84d7512006-05-01 19:59:03 +0200138 .get = generic_pipe_buf_get,
Jens Axboe5274f052006-03-30 15:15:30 +0200139};
140
Jens Axboe912d35f2006-04-26 10:59:21 +0200141static int user_page_pipe_buf_steal(struct pipe_inode_info *pipe,
142 struct pipe_buffer *buf)
143{
Jens Axboe7afa6fd2006-05-01 20:02:33 +0200144 if (!(buf->flags & PIPE_BUF_FLAG_GIFT))
145 return 1;
146
Jens Axboe14328732006-05-03 10:35:26 +0200147 buf->flags |= PIPE_BUF_FLAG_LRU;
Jens Axboe330ab712006-05-02 15:29:57 +0200148 return generic_pipe_buf_steal(pipe, buf);
Jens Axboe912d35f2006-04-26 10:59:21 +0200149}
150
Eric Dumazetd4c3cca2006-12-13 00:34:04 -0800151static const struct pipe_buf_operations user_page_pipe_buf_ops = {
Jens Axboe912d35f2006-04-26 10:59:21 +0200152 .can_merge = 0,
Jens Axboef84d7512006-05-01 19:59:03 +0200153 .map = generic_pipe_buf_map,
154 .unmap = generic_pipe_buf_unmap,
Jens Axboecac36bb02007-06-14 13:10:48 +0200155 .confirm = generic_pipe_buf_confirm,
Jens Axboe912d35f2006-04-26 10:59:21 +0200156 .release = page_cache_pipe_buf_release,
157 .steal = user_page_pipe_buf_steal,
Jens Axboef84d7512006-05-01 19:59:03 +0200158 .get = generic_pipe_buf_get,
Jens Axboe912d35f2006-04-26 10:59:21 +0200159};
160
Jens Axboe932cc6d2007-06-21 13:10:21 +0200161/**
162 * splice_to_pipe - fill passed data into a pipe
163 * @pipe: pipe to fill
164 * @spd: data to fill
165 *
166 * Description:
Randy Dunlap79685b82007-07-27 08:08:51 +0200167 * @spd contains a map of pages and len/offset tuples, along with
Jens Axboe932cc6d2007-06-21 13:10:21 +0200168 * the struct pipe_buf_operations associated with these pages. This
169 * function will link that data to the pipe.
170 *
Jens Axboe83f91352006-04-02 23:05:09 +0200171 */
Jens Axboed6b29d72007-06-04 09:59:47 +0200172ssize_t splice_to_pipe(struct pipe_inode_info *pipe,
173 struct splice_pipe_desc *spd)
Jens Axboe5274f052006-03-30 15:15:30 +0200174{
Jens Axboe00de00b2007-06-15 13:14:22 +0200175 unsigned int spd_pages = spd->nr_pages;
Jens Axboe912d35f2006-04-26 10:59:21 +0200176 int ret, do_wakeup, page_nr;
Jens Axboe5274f052006-03-30 15:15:30 +0200177
178 ret = 0;
179 do_wakeup = 0;
Jens Axboe912d35f2006-04-26 10:59:21 +0200180 page_nr = 0;
Jens Axboe5274f052006-03-30 15:15:30 +0200181
Ingo Molnar3a326a22006-04-10 15:18:35 +0200182 if (pipe->inode)
183 mutex_lock(&pipe->inode->i_mutex);
Jens Axboe5274f052006-03-30 15:15:30 +0200184
Jens Axboe5274f052006-03-30 15:15:30 +0200185 for (;;) {
Ingo Molnar3a326a22006-04-10 15:18:35 +0200186 if (!pipe->readers) {
Jens Axboe5274f052006-03-30 15:15:30 +0200187 send_sig(SIGPIPE, current, 0);
188 if (!ret)
189 ret = -EPIPE;
190 break;
191 }
192
Jens Axboe6f767b02006-04-11 13:53:56 +0200193 if (pipe->nrbufs < PIPE_BUFFERS) {
194 int newbuf = (pipe->curbuf + pipe->nrbufs) & (PIPE_BUFFERS - 1);
Ingo Molnar3a326a22006-04-10 15:18:35 +0200195 struct pipe_buffer *buf = pipe->bufs + newbuf;
Jens Axboe5274f052006-03-30 15:15:30 +0200196
Jens Axboe912d35f2006-04-26 10:59:21 +0200197 buf->page = spd->pages[page_nr];
198 buf->offset = spd->partial[page_nr].offset;
199 buf->len = spd->partial[page_nr].len;
Jens Axboe497f9622007-06-11 12:00:45 +0200200 buf->private = spd->partial[page_nr].private;
Jens Axboe912d35f2006-04-26 10:59:21 +0200201 buf->ops = spd->ops;
Jens Axboe7afa6fd2006-05-01 20:02:33 +0200202 if (spd->flags & SPLICE_F_GIFT)
203 buf->flags |= PIPE_BUF_FLAG_GIFT;
204
Jens Axboe6f767b02006-04-11 13:53:56 +0200205 pipe->nrbufs++;
Jens Axboe912d35f2006-04-26 10:59:21 +0200206 page_nr++;
207 ret += buf->len;
208
Jens Axboe6f767b02006-04-11 13:53:56 +0200209 if (pipe->inode)
210 do_wakeup = 1;
Jens Axboe5274f052006-03-30 15:15:30 +0200211
Jens Axboe912d35f2006-04-26 10:59:21 +0200212 if (!--spd->nr_pages)
Jens Axboe5274f052006-03-30 15:15:30 +0200213 break;
Jens Axboe6f767b02006-04-11 13:53:56 +0200214 if (pipe->nrbufs < PIPE_BUFFERS)
Jens Axboe5274f052006-03-30 15:15:30 +0200215 continue;
216
217 break;
218 }
219
Jens Axboe912d35f2006-04-26 10:59:21 +0200220 if (spd->flags & SPLICE_F_NONBLOCK) {
Linus Torvalds29e35092006-04-02 12:46:35 -0700221 if (!ret)
222 ret = -EAGAIN;
223 break;
224 }
225
Jens Axboe5274f052006-03-30 15:15:30 +0200226 if (signal_pending(current)) {
227 if (!ret)
228 ret = -ERESTARTSYS;
229 break;
230 }
231
232 if (do_wakeup) {
Jens Axboec0bd1f62006-04-10 09:03:32 +0200233 smp_mb();
Ingo Molnar3a326a22006-04-10 15:18:35 +0200234 if (waitqueue_active(&pipe->wait))
235 wake_up_interruptible_sync(&pipe->wait);
236 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
Jens Axboe5274f052006-03-30 15:15:30 +0200237 do_wakeup = 0;
238 }
239
Ingo Molnar3a326a22006-04-10 15:18:35 +0200240 pipe->waiting_writers++;
241 pipe_wait(pipe);
242 pipe->waiting_writers--;
Jens Axboe5274f052006-03-30 15:15:30 +0200243 }
244
Jens Axboe02676e52007-06-15 13:16:13 +0200245 if (pipe->inode) {
Ingo Molnar3a326a22006-04-10 15:18:35 +0200246 mutex_unlock(&pipe->inode->i_mutex);
Jens Axboe5274f052006-03-30 15:15:30 +0200247
Jens Axboe02676e52007-06-15 13:16:13 +0200248 if (do_wakeup) {
249 smp_mb();
250 if (waitqueue_active(&pipe->wait))
251 wake_up_interruptible(&pipe->wait);
252 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
253 }
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 Wu431a48202007-07-19 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 Wu431a48202007-07-19 01:48:05 -0700301 * readahead/allocate the rest and fill in the holes.
Jens Axboeeb207962006-04-27 11:05:22 +0200302 */
Fengguang Wu431a48202007-07-19 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 Wu431a48202007-07-19 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) {
Jens Axboe8191ecd2008-04-10 08:24:25 +0200373 if (TestSetPageLocked(page)) {
374 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 /*
381 * page was truncated, stop here. if this isn't the
382 * first page, we'll just complete what we already
383 * added
384 */
385 if (!page->mapping) {
386 unlock_page(page);
Jens Axboe7480a902006-04-11 13:52:47 +0200387 break;
388 }
389 /*
390 * page was already under io and is now done, great
391 */
392 if (PageUptodate(page)) {
393 unlock_page(page);
394 goto fill_it;
395 }
396
Jens Axboe7480a902006-04-11 13:52:47 +0200397 /*
398 * need to read in the page
399 */
400 error = mapping->a_ops->readpage(in, page);
Jens Axboe7480a902006-04-11 13:52:47 +0200401 if (unlikely(error)) {
Jens Axboeeb207962006-04-27 11:05:22 +0200402 /*
403 * We really should re-lookup the page here,
404 * but it complicates things a lot. Instead
405 * lets just do what we already stored, and
406 * we'll get it the next time we are called.
407 */
Jens Axboe7480a902006-04-11 13:52:47 +0200408 if (error == AOP_TRUNCATED_PAGE)
Jens Axboeeb207962006-04-27 11:05:22 +0200409 error = 0;
410
Jens Axboe7480a902006-04-11 13:52:47 +0200411 break;
412 }
Jens Axboe620a3242007-06-07 09:39:42 +0200413 }
414fill_it:
415 /*
416 * i_size must be checked after PageUptodate.
417 */
418 isize = i_size_read(mapping->host);
419 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
420 if (unlikely(!isize || index > end_index))
421 break;
422
423 /*
424 * if this is the last page, see if we need to shrink
425 * the length and stop
426 */
427 if (end_index == index) {
428 unsigned int plen;
Jens Axboe91ad66e2006-04-19 15:55:10 +0200429
430 /*
Jens Axboe620a3242007-06-07 09:39:42 +0200431 * max good bytes in this page
Jens Axboe91ad66e2006-04-19 15:55:10 +0200432 */
Jens Axboe620a3242007-06-07 09:39:42 +0200433 plen = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
434 if (plen <= loff)
Jens Axboe91ad66e2006-04-19 15:55:10 +0200435 break;
Jens Axboe91ad66e2006-04-19 15:55:10 +0200436
437 /*
Jens Axboe620a3242007-06-07 09:39:42 +0200438 * force quit after adding this page
Jens Axboe91ad66e2006-04-19 15:55:10 +0200439 */
Jens Axboe620a3242007-06-07 09:39:42 +0200440 this_len = min(this_len, plen - loff);
441 len = this_len;
Jens Axboe7480a902006-04-11 13:52:47 +0200442 }
Jens Axboe620a3242007-06-07 09:39:42 +0200443
Jens Axboeeb207962006-04-27 11:05:22 +0200444 partial[page_nr].offset = loff;
445 partial[page_nr].len = this_len;
Jens Axboe82aa5d62006-04-20 13:05:48 +0200446 len -= this_len;
Jens Axboe91ad66e2006-04-19 15:55:10 +0200447 loff = 0;
Jens Axboeeb207962006-04-27 11:05:22 +0200448 spd.nr_pages++;
449 index++;
Jens Axboe5274f052006-03-30 15:15:30 +0200450 }
451
Jens Axboeeb207962006-04-27 11:05:22 +0200452 /*
Hugh Dickins475ecad2007-06-07 09:36:00 +0200453 * Release any pages at the end, if we quit early. 'page_nr' is how far
Jens Axboeeb207962006-04-27 11:05:22 +0200454 * we got, 'nr_pages' is how many pages are in the map.
455 */
456 while (page_nr < nr_pages)
457 page_cache_release(pages[page_nr++]);
Fengguang Wuf4e6b492007-10-16 01:24:33 -0700458 in->f_ra.prev_pos = (loff_t)index << PAGE_CACHE_SHIFT;
Jens Axboeeb207962006-04-27 11:05:22 +0200459
Jens Axboe912d35f2006-04-26 10:59:21 +0200460 if (spd.nr_pages)
Jens Axboe00522fb2006-04-26 14:39:29 +0200461 return splice_to_pipe(pipe, &spd);
Jens Axboe5274f052006-03-30 15:15:30 +0200462
Jens Axboe7480a902006-04-11 13:52:47 +0200463 return error;
Jens Axboe5274f052006-03-30 15:15:30 +0200464}
465
Jens Axboe83f91352006-04-02 23:05:09 +0200466/**
467 * generic_file_splice_read - splice data from file to a pipe
468 * @in: file to splice from
Jens Axboe932cc6d2007-06-21 13:10:21 +0200469 * @ppos: position in @in
Jens Axboe83f91352006-04-02 23:05:09 +0200470 * @pipe: pipe to splice to
471 * @len: number of bytes to splice
472 * @flags: splice modifier flags
473 *
Jens Axboe932cc6d2007-06-21 13:10:21 +0200474 * Description:
475 * Will read pages from given file and fill them into a pipe. Can be
476 * used as long as the address_space operations for the source implements
477 * a readpage() hook.
478 *
Jens Axboe83f91352006-04-02 23:05:09 +0200479 */
Jens Axboecbb7e572006-04-11 14:57:50 +0200480ssize_t generic_file_splice_read(struct file *in, loff_t *ppos,
481 struct pipe_inode_info *pipe, size_t len,
482 unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +0200483{
Jens Axboed366d3982007-06-01 14:54:11 +0200484 loff_t isize, left;
Jens Axboe8191ecd2008-04-10 08:24:25 +0200485 int ret;
Jens Axboed366d3982007-06-01 14:54:11 +0200486
487 isize = i_size_read(in->f_mapping->host);
488 if (unlikely(*ppos >= isize))
489 return 0;
490
491 left = isize - *ppos;
492 if (unlikely(left < len))
493 len = left;
Jens Axboe5274f052006-03-30 15:15:30 +0200494
Jens Axboe8191ecd2008-04-10 08:24:25 +0200495 ret = __generic_file_splice_read(in, ppos, pipe, len, flags);
496 if (ret > 0)
Jens Axboecbb7e572006-04-11 14:57:50 +0200497 *ppos += ret;
Jens Axboe5274f052006-03-30 15:15:30 +0200498
499 return ret;
500}
501
Jens Axboe059a8f32006-04-02 23:06:05 +0200502EXPORT_SYMBOL(generic_file_splice_read);
503
Jens Axboe5274f052006-03-30 15:15:30 +0200504/*
Jens Axboe4f6f0bd2006-04-02 23:04:46 +0200505 * Send 'sd->len' bytes to socket from 'sd->file' at position 'sd->pos'
Jens Axboe016b6612006-04-25 15:42:00 +0200506 * using sendpage(). Return the number of bytes sent.
Jens Axboe5274f052006-03-30 15:15:30 +0200507 */
Jens Axboe76ad4d12006-05-03 10:41:33 +0200508static int pipe_to_sendpage(struct pipe_inode_info *pipe,
Jens Axboe5274f052006-03-30 15:15:30 +0200509 struct pipe_buffer *buf, struct splice_desc *sd)
510{
Jens Axboe6a14b902007-06-14 13:08:55 +0200511 struct file *file = sd->u.file;
Jens Axboe5274f052006-03-30 15:15:30 +0200512 loff_t pos = sd->pos;
Jens Axboef84d7512006-05-01 19:59:03 +0200513 int ret, more;
Jens Axboe5274f052006-03-30 15:15:30 +0200514
Jens Axboecac36bb02007-06-14 13:10:48 +0200515 ret = buf->ops->confirm(pipe, buf);
Jens Axboef84d7512006-05-01 19:59:03 +0200516 if (!ret) {
517 more = (sd->flags & SPLICE_F_MORE) || sd->len < sd->total_len;
Jens Axboe5274f052006-03-30 15:15:30 +0200518
Jens Axboef84d7512006-05-01 19:59:03 +0200519 ret = file->f_op->sendpage(file, buf->page, buf->offset,
520 sd->len, &pos, more);
521 }
Jens Axboe5274f052006-03-30 15:15:30 +0200522
Jens Axboe016b6612006-04-25 15:42:00 +0200523 return ret;
Jens Axboe5274f052006-03-30 15:15:30 +0200524}
525
526/*
527 * This is a little more tricky than the file -> pipe splicing. There are
528 * basically three cases:
529 *
530 * - Destination page already exists in the address space and there
531 * are users of it. For that case we have no other option that
532 * copying the data. Tough luck.
533 * - Destination page already exists in the address space, but there
534 * are no users of it. Make sure it's uptodate, then drop it. Fall
535 * through to last case.
536 * - Destination page does not exist, we can add the pipe page to
537 * the page cache and avoid the copy.
538 *
Jens Axboe83f91352006-04-02 23:05:09 +0200539 * If asked to move pages to the output file (SPLICE_F_MOVE is set in
540 * sd->flags), we attempt to migrate pages from the pipe to the output
541 * file address space page cache. This is possible if no one else has
542 * the pipe page referenced outside of the pipe and page cache. If
543 * SPLICE_F_MOVE isn't set, or we cannot move the page, we simply create
544 * a new page in the output file page cache and fill/dirty that.
Jens Axboe5274f052006-03-30 15:15:30 +0200545 */
Jens Axboe76ad4d12006-05-03 10:41:33 +0200546static int pipe_to_file(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
Jens Axboe5274f052006-03-30 15:15:30 +0200547 struct splice_desc *sd)
548{
Jens Axboe6a14b902007-06-14 13:08:55 +0200549 struct file *file = sd->u.file;
Jens Axboe5274f052006-03-30 15:15:30 +0200550 struct address_space *mapping = file->f_mapping;
Jens Axboe016b6612006-04-25 15:42:00 +0200551 unsigned int offset, this_len;
Jens Axboe5274f052006-03-30 15:15:30 +0200552 struct page *page;
Nick Pigginafddba42007-10-16 01:25:01 -0700553 void *fsdata;
Jens Axboe3e7ee3e2006-04-02 23:11:04 +0200554 int ret;
Jens Axboe5274f052006-03-30 15:15:30 +0200555
556 /*
Jens Axboe49d0b212006-04-10 09:04:41 +0200557 * make sure the data in this buffer is uptodate
Jens Axboe5274f052006-03-30 15:15:30 +0200558 */
Jens Axboecac36bb02007-06-14 13:10:48 +0200559 ret = buf->ops->confirm(pipe, buf);
Jens Axboef84d7512006-05-01 19:59:03 +0200560 if (unlikely(ret))
561 return ret;
Jens Axboe5274f052006-03-30 15:15:30 +0200562
Jens Axboe5274f052006-03-30 15:15:30 +0200563 offset = sd->pos & ~PAGE_CACHE_MASK;
564
Jens Axboe016b6612006-04-25 15:42:00 +0200565 this_len = sd->len;
566 if (this_len + offset > PAGE_CACHE_SIZE)
567 this_len = PAGE_CACHE_SIZE - offset;
568
Nick Pigginafddba42007-10-16 01:25:01 -0700569 ret = pagecache_write_begin(file, mapping, sd->pos, this_len,
570 AOP_FLAG_UNINTERRUPTIBLE, &page, &fsdata);
571 if (unlikely(ret))
572 goto out;
Jens Axboe5274f052006-03-30 15:15:30 +0200573
Jens Axboe0568b402006-05-01 19:50:48 +0200574 if (buf->page != page) {
Jens Axboef84d7512006-05-01 19:59:03 +0200575 /*
576 * Careful, ->map() uses KM_USER0!
577 */
Jens Axboe76ad4d12006-05-03 10:41:33 +0200578 char *src = buf->ops->map(pipe, buf, 1);
Jens Axboef84d7512006-05-01 19:59:03 +0200579 char *dst = kmap_atomic(page, KM_USER1);
Jens Axboe5abc97a2006-03-30 15:16:46 +0200580
Jens Axboe016b6612006-04-25 15:42:00 +0200581 memcpy(dst + offset, src + buf->offset, this_len);
Jens Axboe5abc97a2006-03-30 15:16:46 +0200582 flush_dcache_page(page);
Jens Axboef84d7512006-05-01 19:59:03 +0200583 kunmap_atomic(dst, KM_USER1);
Jens Axboe76ad4d12006-05-03 10:41:33 +0200584 buf->ops->unmap(pipe, buf, src);
Jens Axboe5abc97a2006-03-30 15:16:46 +0200585 }
Nick Pigginafddba42007-10-16 01:25:01 -0700586 ret = pagecache_write_end(file, mapping, sd->pos, this_len, this_len,
587 page, fsdata);
Jens Axboe5274f052006-03-30 15:15:30 +0200588out:
Jens Axboe5274f052006-03-30 15:15:30 +0200589 return ret;
590}
591
Jens Axboe932cc6d2007-06-21 13:10:21 +0200592/**
593 * __splice_from_pipe - splice data from a pipe to given actor
594 * @pipe: pipe to splice from
595 * @sd: information to @actor
596 * @actor: handler that splices the data
597 *
598 * Description:
599 * This function does little more than loop over the pipe and call
600 * @actor to do the actual moving of a single struct pipe_buffer to
601 * the desired destination. See pipe_to_file, pipe_to_sendpage, or
602 * pipe_to_user.
603 *
Jens Axboe83f91352006-04-02 23:05:09 +0200604 */
Jens Axboec66ab6f2007-06-12 21:17:17 +0200605ssize_t __splice_from_pipe(struct pipe_inode_info *pipe, struct splice_desc *sd,
606 splice_actor *actor)
Jens Axboe5274f052006-03-30 15:15:30 +0200607{
Jens Axboe5274f052006-03-30 15:15:30 +0200608 int ret, do_wakeup, err;
Jens Axboe5274f052006-03-30 15:15:30 +0200609
610 ret = 0;
611 do_wakeup = 0;
612
Jens Axboe5274f052006-03-30 15:15:30 +0200613 for (;;) {
Jens Axboe6f767b02006-04-11 13:53:56 +0200614 if (pipe->nrbufs) {
615 struct pipe_buffer *buf = pipe->bufs + pipe->curbuf;
Eric Dumazetd4c3cca2006-12-13 00:34:04 -0800616 const struct pipe_buf_operations *ops = buf->ops;
Jens Axboe5274f052006-03-30 15:15:30 +0200617
Jens Axboec66ab6f2007-06-12 21:17:17 +0200618 sd->len = buf->len;
619 if (sd->len > sd->total_len)
620 sd->len = sd->total_len;
Jens Axboe5274f052006-03-30 15:15:30 +0200621
Jens Axboec66ab6f2007-06-12 21:17:17 +0200622 err = actor(pipe, buf, sd);
Jens Axboe016b6612006-04-25 15:42:00 +0200623 if (err <= 0) {
Jens Axboe5274f052006-03-30 15:15:30 +0200624 if (!ret && err != -ENODATA)
625 ret = err;
626
627 break;
628 }
629
Jens Axboe016b6612006-04-25 15:42:00 +0200630 ret += err;
631 buf->offset += err;
632 buf->len -= err;
633
Jens Axboec66ab6f2007-06-12 21:17:17 +0200634 sd->len -= err;
635 sd->pos += err;
636 sd->total_len -= err;
637 if (sd->len)
Jens Axboe016b6612006-04-25 15:42:00 +0200638 continue;
Ingo Molnar73d62d82006-04-11 13:57:21 +0200639
Jens Axboe5274f052006-03-30 15:15:30 +0200640 if (!buf->len) {
641 buf->ops = NULL;
Ingo Molnar3a326a22006-04-10 15:18:35 +0200642 ops->release(pipe, buf);
Jens Axboe6f767b02006-04-11 13:53:56 +0200643 pipe->curbuf = (pipe->curbuf + 1) & (PIPE_BUFFERS - 1);
644 pipe->nrbufs--;
645 if (pipe->inode)
646 do_wakeup = 1;
Jens Axboe5274f052006-03-30 15:15:30 +0200647 }
648
Jens Axboec66ab6f2007-06-12 21:17:17 +0200649 if (!sd->total_len)
Jens Axboe5274f052006-03-30 15:15:30 +0200650 break;
651 }
652
Jens Axboe6f767b02006-04-11 13:53:56 +0200653 if (pipe->nrbufs)
Jens Axboe5274f052006-03-30 15:15:30 +0200654 continue;
Ingo Molnar3a326a22006-04-10 15:18:35 +0200655 if (!pipe->writers)
Jens Axboe5274f052006-03-30 15:15:30 +0200656 break;
Ingo Molnar3a326a22006-04-10 15:18:35 +0200657 if (!pipe->waiting_writers) {
Jens Axboe5274f052006-03-30 15:15:30 +0200658 if (ret)
659 break;
660 }
661
Jens Axboec66ab6f2007-06-12 21:17:17 +0200662 if (sd->flags & SPLICE_F_NONBLOCK) {
Linus Torvalds29e35092006-04-02 12:46:35 -0700663 if (!ret)
664 ret = -EAGAIN;
665 break;
666 }
667
Jens Axboe5274f052006-03-30 15:15:30 +0200668 if (signal_pending(current)) {
669 if (!ret)
670 ret = -ERESTARTSYS;
671 break;
672 }
673
674 if (do_wakeup) {
Jens Axboec0bd1f62006-04-10 09:03:32 +0200675 smp_mb();
Ingo Molnar3a326a22006-04-10 15:18:35 +0200676 if (waitqueue_active(&pipe->wait))
677 wake_up_interruptible_sync(&pipe->wait);
678 kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
Jens Axboe5274f052006-03-30 15:15:30 +0200679 do_wakeup = 0;
680 }
681
Ingo Molnar3a326a22006-04-10 15:18:35 +0200682 pipe_wait(pipe);
Jens Axboe5274f052006-03-30 15:15:30 +0200683 }
684
Jens Axboe5274f052006-03-30 15:15:30 +0200685 if (do_wakeup) {
Jens Axboec0bd1f62006-04-10 09:03:32 +0200686 smp_mb();
Ingo Molnar3a326a22006-04-10 15:18:35 +0200687 if (waitqueue_active(&pipe->wait))
688 wake_up_interruptible(&pipe->wait);
689 kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
Jens Axboe5274f052006-03-30 15:15:30 +0200690 }
691
Jens Axboe5274f052006-03-30 15:15:30 +0200692 return ret;
Jens Axboe5274f052006-03-30 15:15:30 +0200693}
Mark Fasheh40bee44e2007-03-21 13:11:02 +0100694EXPORT_SYMBOL(__splice_from_pipe);
Jens Axboe5274f052006-03-30 15:15:30 +0200695
Jens Axboe932cc6d2007-06-21 13:10:21 +0200696/**
697 * splice_from_pipe - splice data from a pipe to a file
698 * @pipe: pipe to splice from
699 * @out: file to splice to
700 * @ppos: position in @out
701 * @len: how many bytes to splice
702 * @flags: splice modifier flags
703 * @actor: handler that splices the data
704 *
705 * Description:
706 * See __splice_from_pipe. This function locks the input and output inodes,
707 * otherwise it's identical to __splice_from_pipe().
708 *
709 */
Mark Fasheh6da61802006-10-17 18:43:07 +0200710ssize_t splice_from_pipe(struct pipe_inode_info *pipe, struct file *out,
711 loff_t *ppos, size_t len, unsigned int flags,
712 splice_actor *actor)
713{
714 ssize_t ret;
715 struct inode *inode = out->f_mapping->host;
Jens Axboec66ab6f2007-06-12 21:17:17 +0200716 struct splice_desc sd = {
717 .total_len = len,
718 .flags = flags,
719 .pos = *ppos,
Jens Axboe6a14b902007-06-14 13:08:55 +0200720 .u.file = out,
Jens Axboec66ab6f2007-06-12 21:17:17 +0200721 };
Mark Fasheh6da61802006-10-17 18:43:07 +0200722
723 /*
724 * The actor worker might be calling ->prepare_write and
725 * ->commit_write. Most of the time, these expect i_mutex to
726 * be held. Since this may result in an ABBA deadlock with
727 * pipe->inode, we have to order lock acquiry here.
728 */
729 inode_double_lock(inode, pipe->inode);
Jens Axboec66ab6f2007-06-12 21:17:17 +0200730 ret = __splice_from_pipe(pipe, &sd, actor);
Mark Fasheh6da61802006-10-17 18:43:07 +0200731 inode_double_unlock(inode, pipe->inode);
732
733 return ret;
734}
735
736/**
737 * generic_file_splice_write_nolock - generic_file_splice_write without mutexes
738 * @pipe: pipe info
739 * @out: file to write to
Jens Axboe932cc6d2007-06-21 13:10:21 +0200740 * @ppos: position in @out
Mark Fasheh6da61802006-10-17 18:43:07 +0200741 * @len: number of bytes to splice
742 * @flags: splice modifier flags
743 *
Jens Axboe932cc6d2007-06-21 13:10:21 +0200744 * Description:
745 * Will either move or copy pages (determined by @flags options) from
746 * the given pipe inode to the given file. The caller is responsible
747 * for acquiring i_mutex on both inodes.
Mark Fasheh6da61802006-10-17 18:43:07 +0200748 *
749 */
750ssize_t
751generic_file_splice_write_nolock(struct pipe_inode_info *pipe, struct file *out,
752 loff_t *ppos, size_t len, unsigned int flags)
753{
754 struct address_space *mapping = out->f_mapping;
755 struct inode *inode = mapping->host;
Jens Axboec66ab6f2007-06-12 21:17:17 +0200756 struct splice_desc sd = {
757 .total_len = len,
758 .flags = flags,
759 .pos = *ppos,
Jens Axboe6a14b902007-06-14 13:08:55 +0200760 .u.file = out,
Jens Axboec66ab6f2007-06-12 21:17:17 +0200761 };
Mark Fasheh6da61802006-10-17 18:43:07 +0200762 ssize_t ret;
763 int err;
764
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800765 err = remove_suid(out->f_path.dentry);
Jens Axboe8c34e2d2006-10-17 19:43:22 +0200766 if (unlikely(err))
767 return err;
768
Jens Axboec66ab6f2007-06-12 21:17:17 +0200769 ret = __splice_from_pipe(pipe, &sd, pipe_to_file);
Mark Fasheh6da61802006-10-17 18:43:07 +0200770 if (ret > 0) {
Jens Axboe17ee4f42007-06-15 13:10:37 +0200771 unsigned long nr_pages;
772
Mark Fasheh6da61802006-10-17 18:43:07 +0200773 *ppos += ret;
Jens Axboe17ee4f42007-06-15 13:10:37 +0200774 nr_pages = (ret + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
Mark Fasheh6da61802006-10-17 18:43:07 +0200775
776 /*
777 * If file or inode is SYNC and we actually wrote some data,
778 * sync it.
779 */
780 if (unlikely((out->f_flags & O_SYNC) || IS_SYNC(inode))) {
781 err = generic_osync_inode(inode, mapping,
782 OSYNC_METADATA|OSYNC_DATA);
783
784 if (err)
785 ret = err;
786 }
Jens Axboe17ee4f42007-06-15 13:10:37 +0200787 balance_dirty_pages_ratelimited_nr(mapping, nr_pages);
Mark Fasheh6da61802006-10-17 18:43:07 +0200788 }
789
790 return ret;
791}
792
793EXPORT_SYMBOL(generic_file_splice_write_nolock);
794
Jens Axboe83f91352006-04-02 23:05:09 +0200795/**
796 * generic_file_splice_write - splice data from a pipe to a file
Ingo Molnar3a326a22006-04-10 15:18:35 +0200797 * @pipe: pipe info
Jens Axboe83f91352006-04-02 23:05:09 +0200798 * @out: file to write to
Jens Axboe932cc6d2007-06-21 13:10:21 +0200799 * @ppos: position in @out
Jens Axboe83f91352006-04-02 23:05:09 +0200800 * @len: number of bytes to splice
801 * @flags: splice modifier flags
802 *
Jens Axboe932cc6d2007-06-21 13:10:21 +0200803 * Description:
804 * Will either move or copy pages (determined by @flags options) from
805 * the given pipe inode to the given file.
Jens Axboe83f91352006-04-02 23:05:09 +0200806 *
807 */
Ingo Molnar3a326a22006-04-10 15:18:35 +0200808ssize_t
809generic_file_splice_write(struct pipe_inode_info *pipe, struct file *out,
Jens Axboecbb7e572006-04-11 14:57:50 +0200810 loff_t *ppos, size_t len, unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +0200811{
Jens Axboe4f6f0bd2006-04-02 23:04:46 +0200812 struct address_space *mapping = out->f_mapping;
Jens Axboe8c34e2d2006-10-17 19:43:22 +0200813 struct inode *inode = mapping->host;
Miklos Szeredi7f3d4ee2008-05-07 09:22:39 +0200814 struct splice_desc sd = {
815 .total_len = len,
816 .flags = flags,
817 .pos = *ppos,
818 .u.file = out,
819 };
Ingo Molnar3a326a22006-04-10 15:18:35 +0200820 ssize_t ret;
Jens Axboe8c34e2d2006-10-17 19:43:22 +0200821
Miklos Szeredi7f3d4ee2008-05-07 09:22:39 +0200822 inode_double_lock(inode, pipe->inode);
823 ret = remove_suid(out->f_path.dentry);
824 if (likely(!ret))
825 ret = __splice_from_pipe(pipe, &sd, pipe_to_file);
826 inode_double_unlock(inode, pipe->inode);
Jens Axboea4514eb2006-04-19 15:57:05 +0200827 if (ret > 0) {
Jens Axboe17ee4f42007-06-15 13:10:37 +0200828 unsigned long nr_pages;
829
Jens Axboea4514eb2006-04-19 15:57:05 +0200830 *ppos += ret;
Jens Axboe17ee4f42007-06-15 13:10:37 +0200831 nr_pages = (ret + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
Jens Axboe4f6f0bd2006-04-02 23:04:46 +0200832
Jens Axboea4514eb2006-04-19 15:57:05 +0200833 /*
834 * If file or inode is SYNC and we actually wrote some data,
835 * sync it.
836 */
837 if (unlikely((out->f_flags & O_SYNC) || IS_SYNC(inode))) {
Miklos Szeredi7f3d4ee2008-05-07 09:22:39 +0200838 int err;
839
Jens Axboea4514eb2006-04-19 15:57:05 +0200840 mutex_lock(&inode->i_mutex);
841 err = generic_osync_inode(inode, mapping,
842 OSYNC_METADATA|OSYNC_DATA);
843 mutex_unlock(&inode->i_mutex);
844
845 if (err)
846 ret = err;
847 }
Jens Axboe17ee4f42007-06-15 13:10:37 +0200848 balance_dirty_pages_ratelimited_nr(mapping, nr_pages);
Jens Axboe4f6f0bd2006-04-02 23:04:46 +0200849 }
850
851 return ret;
Jens Axboe5274f052006-03-30 15:15:30 +0200852}
853
Jens Axboe059a8f32006-04-02 23:06:05 +0200854EXPORT_SYMBOL(generic_file_splice_write);
855
Jens Axboe83f91352006-04-02 23:05:09 +0200856/**
857 * generic_splice_sendpage - splice data from a pipe to a socket
Jens Axboe932cc6d2007-06-21 13:10:21 +0200858 * @pipe: pipe to splice from
Jens Axboe83f91352006-04-02 23:05:09 +0200859 * @out: socket to write to
Jens Axboe932cc6d2007-06-21 13:10:21 +0200860 * @ppos: position in @out
Jens Axboe83f91352006-04-02 23:05:09 +0200861 * @len: number of bytes to splice
862 * @flags: splice modifier flags
863 *
Jens Axboe932cc6d2007-06-21 13:10:21 +0200864 * Description:
865 * Will send @len bytes from the pipe to a network socket. No data copying
866 * is involved.
Jens Axboe83f91352006-04-02 23:05:09 +0200867 *
868 */
Ingo Molnar3a326a22006-04-10 15:18:35 +0200869ssize_t generic_splice_sendpage(struct pipe_inode_info *pipe, struct file *out,
Jens Axboecbb7e572006-04-11 14:57:50 +0200870 loff_t *ppos, size_t len, unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +0200871{
Jens Axboe00522fb2006-04-26 14:39:29 +0200872 return splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_sendpage);
Jens Axboe5274f052006-03-30 15:15:30 +0200873}
874
Jens Axboe059a8f32006-04-02 23:06:05 +0200875EXPORT_SYMBOL(generic_splice_sendpage);
Jeff Garzika0f06782006-03-30 23:06:13 -0500876
Jens Axboe83f91352006-04-02 23:05:09 +0200877/*
878 * Attempt to initiate a splice from pipe to file.
879 */
Ingo Molnar3a326a22006-04-10 15:18:35 +0200880static long do_splice_from(struct pipe_inode_info *pipe, struct file *out,
Jens Axboecbb7e572006-04-11 14:57:50 +0200881 loff_t *ppos, size_t len, unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +0200882{
Jens Axboe5274f052006-03-30 15:15:30 +0200883 int ret;
884
Jens Axboe49570e92006-04-11 13:56:09 +0200885 if (unlikely(!out->f_op || !out->f_op->splice_write))
Jens Axboe5274f052006-03-30 15:15:30 +0200886 return -EINVAL;
887
Jens Axboe49570e92006-04-11 13:56:09 +0200888 if (unlikely(!(out->f_mode & FMODE_WRITE)))
Jens Axboe5274f052006-03-30 15:15:30 +0200889 return -EBADF;
890
Jens Axboecbb7e572006-04-11 14:57:50 +0200891 ret = rw_verify_area(WRITE, out, ppos, len);
Jens Axboe5274f052006-03-30 15:15:30 +0200892 if (unlikely(ret < 0))
893 return ret;
894
Jens Axboecbb7e572006-04-11 14:57:50 +0200895 return out->f_op->splice_write(pipe, out, ppos, len, flags);
Jens Axboe5274f052006-03-30 15:15:30 +0200896}
897
Jens Axboe83f91352006-04-02 23:05:09 +0200898/*
899 * Attempt to initiate a splice from a file to a pipe.
900 */
Jens Axboecbb7e572006-04-11 14:57:50 +0200901static long do_splice_to(struct file *in, loff_t *ppos,
902 struct pipe_inode_info *pipe, size_t len,
903 unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +0200904{
Jens Axboe5274f052006-03-30 15:15:30 +0200905 int ret;
906
Jens Axboe49570e92006-04-11 13:56:09 +0200907 if (unlikely(!in->f_op || !in->f_op->splice_read))
Jens Axboe5274f052006-03-30 15:15:30 +0200908 return -EINVAL;
909
Jens Axboe49570e92006-04-11 13:56:09 +0200910 if (unlikely(!(in->f_mode & FMODE_READ)))
Jens Axboe5274f052006-03-30 15:15:30 +0200911 return -EBADF;
912
Jens Axboecbb7e572006-04-11 14:57:50 +0200913 ret = rw_verify_area(READ, in, ppos, len);
Jens Axboe5274f052006-03-30 15:15:30 +0200914 if (unlikely(ret < 0))
915 return ret;
916
Jens Axboecbb7e572006-04-11 14:57:50 +0200917 return in->f_op->splice_read(in, ppos, pipe, len, flags);
Jens Axboe5274f052006-03-30 15:15:30 +0200918}
919
Jens Axboe932cc6d2007-06-21 13:10:21 +0200920/**
921 * splice_direct_to_actor - splices data directly between two non-pipes
922 * @in: file to splice from
923 * @sd: actor information on where to splice to
924 * @actor: handles the data splicing
925 *
926 * Description:
927 * This is a special case helper to splice directly between two
928 * points, without requiring an explicit pipe. Internally an allocated
Randy Dunlap79685b82007-07-27 08:08:51 +0200929 * pipe is cached in the process, and reused during the lifetime of
Jens Axboe932cc6d2007-06-21 13:10:21 +0200930 * that process.
931 *
Jens Axboec66ab6f2007-06-12 21:17:17 +0200932 */
933ssize_t splice_direct_to_actor(struct file *in, struct splice_desc *sd,
934 splice_direct_actor *actor)
Jens Axboeb92ce552006-04-11 13:52:07 +0200935{
936 struct pipe_inode_info *pipe;
937 long ret, bytes;
938 umode_t i_mode;
Jens Axboec66ab6f2007-06-12 21:17:17 +0200939 size_t len;
940 int i, flags;
Jens Axboeb92ce552006-04-11 13:52:07 +0200941
942 /*
943 * We require the input being a regular file, as we don't want to
944 * randomly drop data for eg socket -> socket splicing. Use the
945 * piped splicing for that!
946 */
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800947 i_mode = in->f_path.dentry->d_inode->i_mode;
Jens Axboeb92ce552006-04-11 13:52:07 +0200948 if (unlikely(!S_ISREG(i_mode) && !S_ISBLK(i_mode)))
949 return -EINVAL;
950
951 /*
952 * neither in nor out is a pipe, setup an internal pipe attached to
953 * 'out' and transfer the wanted data from 'in' to 'out' through that
954 */
955 pipe = current->splice_pipe;
Jens Axboe49570e92006-04-11 13:56:09 +0200956 if (unlikely(!pipe)) {
Jens Axboeb92ce552006-04-11 13:52:07 +0200957 pipe = alloc_pipe_info(NULL);
958 if (!pipe)
959 return -ENOMEM;
960
961 /*
962 * We don't have an immediate reader, but we'll read the stuff
Jens Axboe00522fb2006-04-26 14:39:29 +0200963 * out of the pipe right after the splice_to_pipe(). So set
Jens Axboeb92ce552006-04-11 13:52:07 +0200964 * PIPE_READERS appropriately.
965 */
966 pipe->readers = 1;
967
968 current->splice_pipe = pipe;
969 }
970
971 /*
Ingo Molnar73d62d82006-04-11 13:57:21 +0200972 * Do the splice.
Jens Axboeb92ce552006-04-11 13:52:07 +0200973 */
974 ret = 0;
975 bytes = 0;
Jens Axboec66ab6f2007-06-12 21:17:17 +0200976 len = sd->total_len;
977 flags = sd->flags;
978
979 /*
980 * Don't block on output, we have to drain the direct pipe.
981 */
982 sd->flags &= ~SPLICE_F_NONBLOCK;
Jens Axboeb92ce552006-04-11 13:52:07 +0200983
984 while (len) {
Jens Axboe51a92c02007-07-13 14:11:43 +0200985 size_t read_len;
Jens Axboebcd4f3a2007-07-16 14:41:49 +0200986 loff_t pos = sd->pos;
Jens Axboeb92ce552006-04-11 13:52:07 +0200987
Jens Axboebcd4f3a2007-07-16 14:41:49 +0200988 ret = do_splice_to(in, &pos, pipe, len, flags);
Jens Axboe51a92c02007-07-13 14:11:43 +0200989 if (unlikely(ret <= 0))
Jens Axboeb92ce552006-04-11 13:52:07 +0200990 goto out_release;
991
992 read_len = ret;
Jens Axboec66ab6f2007-06-12 21:17:17 +0200993 sd->total_len = read_len;
Jens Axboeb92ce552006-04-11 13:52:07 +0200994
995 /*
996 * NOTE: nonblocking mode only applies to the input. We
997 * must not do the output in nonblocking mode as then we
998 * could get stuck data in the internal pipe:
999 */
Jens Axboec66ab6f2007-06-12 21:17:17 +02001000 ret = actor(pipe, sd);
Jens Axboe51a92c02007-07-13 14:11:43 +02001001 if (unlikely(ret <= 0))
Jens Axboeb92ce552006-04-11 13:52:07 +02001002 goto out_release;
1003
1004 bytes += ret;
1005 len -= ret;
Jens Axboebcd4f3a2007-07-16 14:41:49 +02001006 sd->pos = pos;
Jens Axboeb92ce552006-04-11 13:52:07 +02001007
Jens Axboe51a92c02007-07-13 14:11:43 +02001008 if (ret < read_len)
1009 goto out_release;
Jens Axboeb92ce552006-04-11 13:52:07 +02001010 }
1011
Jens Axboe9e971982008-01-29 21:05:57 +01001012done:
Jens Axboeb92ce552006-04-11 13:52:07 +02001013 pipe->nrbufs = pipe->curbuf = 0;
Jens Axboe80848702008-01-30 12:24:48 +01001014 file_accessed(in);
Jens Axboeb92ce552006-04-11 13:52:07 +02001015 return bytes;
1016
1017out_release:
1018 /*
1019 * If we did an incomplete transfer we must release
1020 * the pipe buffers in question:
1021 */
1022 for (i = 0; i < PIPE_BUFFERS; i++) {
1023 struct pipe_buffer *buf = pipe->bufs + i;
1024
1025 if (buf->ops) {
1026 buf->ops->release(pipe, buf);
1027 buf->ops = NULL;
1028 }
1029 }
Jens Axboeb92ce552006-04-11 13:52:07 +02001030
Jens Axboe9e971982008-01-29 21:05:57 +01001031 if (!bytes)
1032 bytes = ret;
Jens Axboeb92ce552006-04-11 13:52:07 +02001033
Jens Axboe9e971982008-01-29 21:05:57 +01001034 goto done;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001035}
1036EXPORT_SYMBOL(splice_direct_to_actor);
1037
1038static int direct_splice_actor(struct pipe_inode_info *pipe,
1039 struct splice_desc *sd)
1040{
Jens Axboe6a14b902007-06-14 13:08:55 +02001041 struct file *file = sd->u.file;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001042
1043 return do_splice_from(pipe, file, &sd->pos, sd->total_len, sd->flags);
1044}
1045
Jens Axboe932cc6d2007-06-21 13:10:21 +02001046/**
1047 * do_splice_direct - splices data directly between two files
1048 * @in: file to splice from
1049 * @ppos: input file offset
1050 * @out: file to splice to
1051 * @len: number of bytes to splice
1052 * @flags: splice modifier flags
1053 *
1054 * Description:
1055 * For use by do_sendfile(). splice can easily emulate sendfile, but
1056 * doing it in the application would incur an extra system call
1057 * (splice in + splice out, as compared to just sendfile()). So this helper
1058 * can splice directly through a process-private pipe.
1059 *
1060 */
Jens Axboec66ab6f2007-06-12 21:17:17 +02001061long do_splice_direct(struct file *in, loff_t *ppos, struct file *out,
1062 size_t len, unsigned int flags)
1063{
1064 struct splice_desc sd = {
1065 .len = len,
1066 .total_len = len,
1067 .flags = flags,
1068 .pos = *ppos,
Jens Axboe6a14b902007-06-14 13:08:55 +02001069 .u.file = out,
Jens Axboec66ab6f2007-06-12 21:17:17 +02001070 };
Jens Axboe51a92c02007-07-13 14:11:43 +02001071 long ret;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001072
1073 ret = splice_direct_to_actor(in, &sd, direct_splice_actor);
Jens Axboe51a92c02007-07-13 14:11:43 +02001074 if (ret > 0)
Jens Axboe75065ff2008-05-08 14:06:19 +02001075 *ppos += ret;
Jens Axboe51a92c02007-07-13 14:11:43 +02001076
Jens Axboec66ab6f2007-06-12 21:17:17 +02001077 return ret;
Jens Axboeb92ce552006-04-11 13:52:07 +02001078}
1079
Jens Axboe83f91352006-04-02 23:05:09 +02001080/*
Jens Axboeddac0d32006-11-04 12:49:32 +01001081 * After the inode slimming patch, i_pipe/i_bdev/i_cdev share the same
1082 * location, so checking ->i_pipe is not enough to verify that this is a
1083 * pipe.
1084 */
1085static inline struct pipe_inode_info *pipe_info(struct inode *inode)
1086{
1087 if (S_ISFIFO(inode->i_mode))
1088 return inode->i_pipe;
1089
1090 return NULL;
1091}
1092
1093/*
Jens Axboe83f91352006-04-02 23:05:09 +02001094 * Determine where to splice to/from.
1095 */
Ingo Molnar529565d2006-04-10 15:18:58 +02001096static long do_splice(struct file *in, loff_t __user *off_in,
1097 struct file *out, loff_t __user *off_out,
1098 size_t len, unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +02001099{
Ingo Molnar3a326a22006-04-10 15:18:35 +02001100 struct pipe_inode_info *pipe;
Jens Axboecbb7e572006-04-11 14:57:50 +02001101 loff_t offset, *off;
Jens Axboea4514eb2006-04-19 15:57:05 +02001102 long ret;
Jens Axboe5274f052006-03-30 15:15:30 +02001103
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001104 pipe = pipe_info(in->f_path.dentry->d_inode);
Ingo Molnar529565d2006-04-10 15:18:58 +02001105 if (pipe) {
1106 if (off_in)
1107 return -ESPIPE;
Jens Axboeb92ce552006-04-11 13:52:07 +02001108 if (off_out) {
1109 if (out->f_op->llseek == no_llseek)
1110 return -EINVAL;
Jens Axboecbb7e572006-04-11 14:57:50 +02001111 if (copy_from_user(&offset, off_out, sizeof(loff_t)))
Jens Axboeb92ce552006-04-11 13:52:07 +02001112 return -EFAULT;
Jens Axboecbb7e572006-04-11 14:57:50 +02001113 off = &offset;
1114 } else
1115 off = &out->f_pos;
Ingo Molnar529565d2006-04-10 15:18:58 +02001116
Jens Axboea4514eb2006-04-19 15:57:05 +02001117 ret = do_splice_from(pipe, out, off, len, flags);
1118
1119 if (off_out && copy_to_user(off_out, off, sizeof(loff_t)))
1120 ret = -EFAULT;
1121
1122 return ret;
Ingo Molnar529565d2006-04-10 15:18:58 +02001123 }
Jens Axboe5274f052006-03-30 15:15:30 +02001124
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001125 pipe = pipe_info(out->f_path.dentry->d_inode);
Ingo Molnar529565d2006-04-10 15:18:58 +02001126 if (pipe) {
1127 if (off_out)
1128 return -ESPIPE;
Jens Axboeb92ce552006-04-11 13:52:07 +02001129 if (off_in) {
1130 if (in->f_op->llseek == no_llseek)
1131 return -EINVAL;
Jens Axboecbb7e572006-04-11 14:57:50 +02001132 if (copy_from_user(&offset, off_in, sizeof(loff_t)))
Jens Axboeb92ce552006-04-11 13:52:07 +02001133 return -EFAULT;
Jens Axboecbb7e572006-04-11 14:57:50 +02001134 off = &offset;
1135 } else
1136 off = &in->f_pos;
Ingo Molnar529565d2006-04-10 15:18:58 +02001137
Jens Axboea4514eb2006-04-19 15:57:05 +02001138 ret = do_splice_to(in, off, pipe, len, flags);
1139
1140 if (off_in && copy_to_user(off_in, off, sizeof(loff_t)))
1141 ret = -EFAULT;
1142
1143 return ret;
Ingo Molnar529565d2006-04-10 15:18:58 +02001144 }
Jens Axboe5274f052006-03-30 15:15:30 +02001145
1146 return -EINVAL;
1147}
1148
Jens Axboe912d35f2006-04-26 10:59:21 +02001149/*
Linus Torvalds75723952007-10-01 13:17:28 -07001150 * Do a copy-from-user while holding the mmap_semaphore for reading, in a
1151 * manner safe from deadlocking with simultaneous mmap() (grabbing mmap_sem
1152 * for writing) and page faulting on the user memory pointed to by src.
1153 * This assumes that we will very rarely hit the partial != 0 path, or this
1154 * will not be a win.
1155 */
1156static int copy_from_user_mmap_sem(void *dst, const void __user *src, size_t n)
1157{
1158 int partial;
1159
Jens Axboe88119302008-02-08 08:49:14 -08001160 if (!access_ok(VERIFY_READ, src, n))
1161 return -EFAULT;
1162
Linus Torvalds75723952007-10-01 13:17:28 -07001163 pagefault_disable();
1164 partial = __copy_from_user_inatomic(dst, src, n);
1165 pagefault_enable();
1166
1167 /*
1168 * Didn't copy everything, drop the mmap_sem and do a faulting copy
1169 */
1170 if (unlikely(partial)) {
1171 up_read(&current->mm->mmap_sem);
1172 partial = copy_from_user(dst, src, n);
1173 down_read(&current->mm->mmap_sem);
1174 }
1175
1176 return partial;
1177}
1178
1179/*
Jens Axboe912d35f2006-04-26 10:59:21 +02001180 * Map an iov into an array of pages and offset/length tupples. With the
1181 * partial_page structure, we can map several non-contiguous ranges into
1182 * our ones pages[] map instead of splitting that operation into pieces.
1183 * Could easily be exported as a generic helper for other users, in which
1184 * case one would probably want to add a 'max_nr_pages' parameter as well.
1185 */
1186static int get_iovec_page_array(const struct iovec __user *iov,
1187 unsigned int nr_vecs, struct page **pages,
Jens Axboe7afa6fd2006-05-01 20:02:33 +02001188 struct partial_page *partial, int aligned)
Jens Axboe912d35f2006-04-26 10:59:21 +02001189{
1190 int buffers = 0, error = 0;
1191
Jens Axboe912d35f2006-04-26 10:59:21 +02001192 down_read(&current->mm->mmap_sem);
1193
1194 while (nr_vecs) {
1195 unsigned long off, npages;
Linus Torvalds75723952007-10-01 13:17:28 -07001196 struct iovec entry;
Jens Axboe912d35f2006-04-26 10:59:21 +02001197 void __user *base;
1198 size_t len;
1199 int i;
1200
Linus Torvalds75723952007-10-01 13:17:28 -07001201 error = -EFAULT;
1202 if (copy_from_user_mmap_sem(&entry, iov, sizeof(entry)))
Jens Axboe912d35f2006-04-26 10:59:21 +02001203 break;
Linus Torvalds75723952007-10-01 13:17:28 -07001204
1205 base = entry.iov_base;
1206 len = entry.iov_len;
Jens Axboe912d35f2006-04-26 10:59:21 +02001207
1208 /*
1209 * Sanity check this iovec. 0 read succeeds.
1210 */
Linus Torvalds75723952007-10-01 13:17:28 -07001211 error = 0;
Jens Axboe912d35f2006-04-26 10:59:21 +02001212 if (unlikely(!len))
1213 break;
1214 error = -EFAULT;
Bastian Blank712a30e2008-02-10 16:47:57 +02001215 if (!access_ok(VERIFY_READ, base, len))
Jens Axboe912d35f2006-04-26 10:59:21 +02001216 break;
1217
1218 /*
1219 * Get this base offset and number of pages, then map
1220 * in the user pages.
1221 */
1222 off = (unsigned long) base & ~PAGE_MASK;
Jens Axboe7afa6fd2006-05-01 20:02:33 +02001223
1224 /*
1225 * If asked for alignment, the offset must be zero and the
1226 * length a multiple of the PAGE_SIZE.
1227 */
1228 error = -EINVAL;
1229 if (aligned && (off || len & ~PAGE_MASK))
1230 break;
1231
Jens Axboe912d35f2006-04-26 10:59:21 +02001232 npages = (off + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1233 if (npages > PIPE_BUFFERS - buffers)
1234 npages = PIPE_BUFFERS - buffers;
1235
1236 error = get_user_pages(current, current->mm,
1237 (unsigned long) base, npages, 0, 0,
1238 &pages[buffers], NULL);
1239
1240 if (unlikely(error <= 0))
1241 break;
1242
1243 /*
1244 * Fill this contiguous range into the partial page map.
1245 */
1246 for (i = 0; i < error; i++) {
Jens Axboe75914892006-05-02 12:57:18 +02001247 const int plen = min_t(size_t, len, PAGE_SIZE - off);
Jens Axboe912d35f2006-04-26 10:59:21 +02001248
1249 partial[buffers].offset = off;
1250 partial[buffers].len = plen;
1251
1252 off = 0;
1253 len -= plen;
1254 buffers++;
1255 }
1256
1257 /*
1258 * We didn't complete this iov, stop here since it probably
1259 * means we have to move some of this into a pipe to
1260 * be able to continue.
1261 */
1262 if (len)
1263 break;
1264
1265 /*
1266 * Don't continue if we mapped fewer pages than we asked for,
1267 * or if we mapped the max number of pages that we have
1268 * room for.
1269 */
1270 if (error < npages || buffers == PIPE_BUFFERS)
1271 break;
1272
1273 nr_vecs--;
1274 iov++;
1275 }
1276
1277 up_read(&current->mm->mmap_sem);
1278
1279 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 Axboecac36bb02007-06-14 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
1343 if (pipe->inode)
1344 mutex_lock(&pipe->inode->i_mutex);
1345
1346 error = ret = 0;
1347 while (nr_segs) {
1348 void __user *base;
1349 size_t len;
1350
1351 /*
1352 * Get user address base and length for this iovec.
1353 */
1354 error = get_user(base, &iov->iov_base);
1355 if (unlikely(error))
1356 break;
1357 error = get_user(len, &iov->iov_len);
1358 if (unlikely(error))
1359 break;
1360
1361 /*
1362 * Sanity check this iovec. 0 read succeeds.
1363 */
1364 if (unlikely(!len))
1365 break;
1366 if (unlikely(!base)) {
1367 error = -EFAULT;
1368 break;
1369 }
1370
Jens Axboe88119302008-02-08 08:49:14 -08001371 if (unlikely(!access_ok(VERIFY_WRITE, base, len))) {
1372 error = -EFAULT;
1373 break;
1374 }
1375
Jens Axboe6a14b902007-06-14 13:08:55 +02001376 sd.len = 0;
1377 sd.total_len = len;
1378 sd.flags = flags;
1379 sd.u.userptr = base;
1380 sd.pos = 0;
1381
1382 size = __splice_from_pipe(pipe, &sd, pipe_to_user);
1383 if (size < 0) {
1384 if (!ret)
1385 ret = size;
1386
1387 break;
1388 }
1389
1390 ret += size;
1391
1392 if (size < len)
1393 break;
1394
1395 nr_segs--;
1396 iov++;
1397 }
1398
1399 if (pipe->inode)
1400 mutex_unlock(&pipe->inode->i_mutex);
1401
1402 if (!ret)
1403 ret = error;
1404
1405 return ret;
1406}
1407
Jens Axboe912d35f2006-04-26 10:59:21 +02001408/*
1409 * vmsplice splices a user address range into a pipe. It can be thought of
1410 * as splice-from-memory, where the regular splice is splice-from-file (or
1411 * to file). In both cases the output is a pipe, naturally.
Jens Axboe912d35f2006-04-26 10:59:21 +02001412 */
Jens Axboe6a14b902007-06-14 13:08:55 +02001413static long vmsplice_to_pipe(struct file *file, const struct iovec __user *iov,
1414 unsigned long nr_segs, unsigned int flags)
Jens Axboe912d35f2006-04-26 10:59:21 +02001415{
Jens Axboeddac0d32006-11-04 12:49:32 +01001416 struct pipe_inode_info *pipe;
Jens Axboe912d35f2006-04-26 10:59:21 +02001417 struct page *pages[PIPE_BUFFERS];
1418 struct partial_page partial[PIPE_BUFFERS];
1419 struct splice_pipe_desc spd = {
1420 .pages = pages,
1421 .partial = partial,
1422 .flags = flags,
1423 .ops = &user_page_pipe_buf_ops,
Jens Axboebbdfc2f2007-11-06 23:29:47 -08001424 .spd_release = spd_release_page,
Jens Axboe912d35f2006-04-26 10:59:21 +02001425 };
1426
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001427 pipe = pipe_info(file->f_path.dentry->d_inode);
Jens Axboeddac0d32006-11-04 12:49:32 +01001428 if (!pipe)
Jens Axboe912d35f2006-04-26 10:59:21 +02001429 return -EBADF;
Jens Axboe912d35f2006-04-26 10:59:21 +02001430
Jens Axboe7afa6fd2006-05-01 20:02:33 +02001431 spd.nr_pages = get_iovec_page_array(iov, nr_segs, pages, partial,
1432 flags & SPLICE_F_GIFT);
Jens Axboe912d35f2006-04-26 10:59:21 +02001433 if (spd.nr_pages <= 0)
1434 return spd.nr_pages;
1435
Jens Axboe00522fb2006-04-26 14:39:29 +02001436 return splice_to_pipe(pipe, &spd);
Jens Axboe912d35f2006-04-26 10:59:21 +02001437}
1438
Jens Axboe6a14b902007-06-14 13:08:55 +02001439/*
1440 * Note that vmsplice only really supports true splicing _from_ user memory
1441 * to a pipe, not the other way around. Splicing from user memory is a simple
1442 * operation that can be supported without any funky alignment restrictions
1443 * or nasty vm tricks. We simply map in the user memory and fill them into
1444 * a pipe. The reverse isn't quite as easy, though. There are two possible
1445 * solutions for that:
1446 *
1447 * - memcpy() the data internally, at which point we might as well just
1448 * do a regular read() on the buffer anyway.
1449 * - Lots of nasty vm tricks, that are neither fast nor flexible (it
1450 * has restriction limitations on both ends of the pipe).
1451 *
1452 * Currently we punt and implement it as a normal copy, see pipe_to_user().
1453 *
1454 */
Jens Axboe912d35f2006-04-26 10:59:21 +02001455asmlinkage long sys_vmsplice(int fd, const struct iovec __user *iov,
1456 unsigned long nr_segs, unsigned int flags)
1457{
1458 struct file *file;
1459 long error;
1460 int fput;
1461
Jens Axboe6a14b902007-06-14 13:08:55 +02001462 if (unlikely(nr_segs > UIO_MAXIOV))
1463 return -EINVAL;
1464 else if (unlikely(!nr_segs))
1465 return 0;
1466
Jens Axboe912d35f2006-04-26 10:59:21 +02001467 error = -EBADF;
1468 file = fget_light(fd, &fput);
1469 if (file) {
1470 if (file->f_mode & FMODE_WRITE)
Jens Axboe6a14b902007-06-14 13:08:55 +02001471 error = vmsplice_to_pipe(file, iov, nr_segs, flags);
1472 else if (file->f_mode & FMODE_READ)
1473 error = vmsplice_to_user(file, iov, nr_segs, flags);
Jens Axboe912d35f2006-04-26 10:59:21 +02001474
1475 fput_light(file, fput);
1476 }
1477
1478 return error;
1479}
1480
Ingo Molnar529565d2006-04-10 15:18:58 +02001481asmlinkage long sys_splice(int fd_in, loff_t __user *off_in,
1482 int fd_out, loff_t __user *off_out,
1483 size_t len, unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +02001484{
1485 long error;
1486 struct file *in, *out;
1487 int fput_in, fput_out;
1488
1489 if (unlikely(!len))
1490 return 0;
1491
1492 error = -EBADF;
Ingo Molnar529565d2006-04-10 15:18:58 +02001493 in = fget_light(fd_in, &fput_in);
Jens Axboe5274f052006-03-30 15:15:30 +02001494 if (in) {
1495 if (in->f_mode & FMODE_READ) {
Ingo Molnar529565d2006-04-10 15:18:58 +02001496 out = fget_light(fd_out, &fput_out);
Jens Axboe5274f052006-03-30 15:15:30 +02001497 if (out) {
1498 if (out->f_mode & FMODE_WRITE)
Ingo Molnar529565d2006-04-10 15:18:58 +02001499 error = do_splice(in, off_in,
1500 out, off_out,
1501 len, flags);
Jens Axboe5274f052006-03-30 15:15:30 +02001502 fput_light(out, fput_out);
1503 }
1504 }
1505
1506 fput_light(in, fput_in);
1507 }
1508
1509 return error;
1510}
Jens Axboe70524492006-04-11 15:51:17 +02001511
1512/*
Jens Axboeaadd06e2006-07-10 11:00:01 +02001513 * Make sure there's data to read. Wait for input if we can, otherwise
1514 * return an appropriate error.
1515 */
1516static int link_ipipe_prep(struct pipe_inode_info *pipe, unsigned int flags)
1517{
1518 int ret;
1519
1520 /*
1521 * Check ->nrbufs without the inode lock first. This function
1522 * is speculative anyways, so missing one is ok.
1523 */
1524 if (pipe->nrbufs)
1525 return 0;
1526
1527 ret = 0;
1528 mutex_lock(&pipe->inode->i_mutex);
1529
1530 while (!pipe->nrbufs) {
1531 if (signal_pending(current)) {
1532 ret = -ERESTARTSYS;
1533 break;
1534 }
1535 if (!pipe->writers)
1536 break;
1537 if (!pipe->waiting_writers) {
1538 if (flags & SPLICE_F_NONBLOCK) {
1539 ret = -EAGAIN;
1540 break;
1541 }
1542 }
1543 pipe_wait(pipe);
1544 }
1545
1546 mutex_unlock(&pipe->inode->i_mutex);
1547 return ret;
1548}
1549
1550/*
1551 * Make sure there's writeable room. Wait for room if we can, otherwise
1552 * return an appropriate error.
1553 */
1554static int link_opipe_prep(struct pipe_inode_info *pipe, unsigned int flags)
1555{
1556 int ret;
1557
1558 /*
1559 * Check ->nrbufs without the inode lock first. This function
1560 * is speculative anyways, so missing one is ok.
1561 */
1562 if (pipe->nrbufs < PIPE_BUFFERS)
1563 return 0;
1564
1565 ret = 0;
1566 mutex_lock(&pipe->inode->i_mutex);
1567
1568 while (pipe->nrbufs >= PIPE_BUFFERS) {
1569 if (!pipe->readers) {
1570 send_sig(SIGPIPE, current, 0);
1571 ret = -EPIPE;
1572 break;
1573 }
1574 if (flags & SPLICE_F_NONBLOCK) {
1575 ret = -EAGAIN;
1576 break;
1577 }
1578 if (signal_pending(current)) {
1579 ret = -ERESTARTSYS;
1580 break;
1581 }
1582 pipe->waiting_writers++;
1583 pipe_wait(pipe);
1584 pipe->waiting_writers--;
1585 }
1586
1587 mutex_unlock(&pipe->inode->i_mutex);
1588 return ret;
1589}
1590
1591/*
Jens Axboe70524492006-04-11 15:51:17 +02001592 * Link contents of ipipe to opipe.
1593 */
1594static int link_pipe(struct pipe_inode_info *ipipe,
1595 struct pipe_inode_info *opipe,
1596 size_t len, unsigned int flags)
1597{
1598 struct pipe_buffer *ibuf, *obuf;
Jens Axboeaadd06e2006-07-10 11:00:01 +02001599 int ret = 0, i = 0, nbuf;
Jens Axboe70524492006-04-11 15:51:17 +02001600
1601 /*
1602 * Potential ABBA deadlock, work around it by ordering lock
1603 * grabbing by inode address. Otherwise two different processes
1604 * could deadlock (one doing tee from A -> B, the other from B -> A).
1605 */
Mark Fasheh62752ee2006-10-17 10:31:38 +02001606 inode_double_lock(ipipe->inode, opipe->inode);
Jens Axboe70524492006-04-11 15:51:17 +02001607
Jens Axboeaadd06e2006-07-10 11:00:01 +02001608 do {
Jens Axboe70524492006-04-11 15:51:17 +02001609 if (!opipe->readers) {
1610 send_sig(SIGPIPE, current, 0);
1611 if (!ret)
1612 ret = -EPIPE;
1613 break;
1614 }
Jens Axboe70524492006-04-11 15:51:17 +02001615
1616 /*
Jens Axboeaadd06e2006-07-10 11:00:01 +02001617 * If we have iterated all input buffers or ran out of
1618 * output room, break.
Jens Axboe70524492006-04-11 15:51:17 +02001619 */
Jens Axboeaadd06e2006-07-10 11:00:01 +02001620 if (i >= ipipe->nrbufs || opipe->nrbufs >= PIPE_BUFFERS)
Jens Axboe70524492006-04-11 15:51:17 +02001621 break;
Jens Axboeaadd06e2006-07-10 11:00:01 +02001622
1623 ibuf = ipipe->bufs + ((ipipe->curbuf + i) & (PIPE_BUFFERS - 1));
1624 nbuf = (opipe->curbuf + opipe->nrbufs) & (PIPE_BUFFERS - 1);
1625
Jens Axboe2a27250e2006-04-19 15:56:40 +02001626 /*
Jens Axboeaadd06e2006-07-10 11:00:01 +02001627 * Get a reference to this pipe buffer,
1628 * so we can copy the contents over.
Jens Axboe2a27250e2006-04-19 15:56:40 +02001629 */
Jens Axboeaadd06e2006-07-10 11:00:01 +02001630 ibuf->ops->get(ipipe, ibuf);
Jens Axboe70524492006-04-11 15:51:17 +02001631
Jens Axboeaadd06e2006-07-10 11:00:01 +02001632 obuf = opipe->bufs + nbuf;
1633 *obuf = *ibuf;
Jens Axboe70524492006-04-11 15:51:17 +02001634
Jens Axboeaadd06e2006-07-10 11:00:01 +02001635 /*
1636 * Don't inherit the gift flag, we need to
1637 * prevent multiple steals of this page.
1638 */
1639 obuf->flags &= ~PIPE_BUF_FLAG_GIFT;
1640
1641 if (obuf->len > len)
1642 obuf->len = len;
1643
1644 opipe->nrbufs++;
1645 ret += obuf->len;
1646 len -= obuf->len;
1647 i++;
1648 } while (len);
Jens Axboe70524492006-04-11 15:51:17 +02001649
Jens Axboe02cf01a2008-02-20 10:34:51 +01001650 /*
1651 * return EAGAIN if we have the potential of some data in the
1652 * future, otherwise just return 0
1653 */
1654 if (!ret && ipipe->waiting_writers && (flags & SPLICE_F_NONBLOCK))
1655 ret = -EAGAIN;
1656
Mark Fasheh62752ee2006-10-17 10:31:38 +02001657 inode_double_unlock(ipipe->inode, opipe->inode);
Jens Axboe70524492006-04-11 15:51:17 +02001658
Jens Axboeaadd06e2006-07-10 11:00:01 +02001659 /*
1660 * If we put data in the output pipe, wakeup any potential readers.
1661 */
1662 if (ret > 0) {
Jens Axboe70524492006-04-11 15:51:17 +02001663 smp_mb();
1664 if (waitqueue_active(&opipe->wait))
1665 wake_up_interruptible(&opipe->wait);
1666 kill_fasync(&opipe->fasync_readers, SIGIO, POLL_IN);
1667 }
1668
1669 return ret;
1670}
1671
1672/*
1673 * This is a tee(1) implementation that works on pipes. It doesn't copy
1674 * any data, it simply references the 'in' pages on the 'out' pipe.
1675 * The 'flags' used are the SPLICE_F_* variants, currently the only
1676 * applicable one is SPLICE_F_NONBLOCK.
1677 */
1678static long do_tee(struct file *in, struct file *out, size_t len,
1679 unsigned int flags)
1680{
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001681 struct pipe_inode_info *ipipe = pipe_info(in->f_path.dentry->d_inode);
1682 struct pipe_inode_info *opipe = pipe_info(out->f_path.dentry->d_inode);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001683 int ret = -EINVAL;
Jens Axboe70524492006-04-11 15:51:17 +02001684
1685 /*
Jens Axboeaadd06e2006-07-10 11:00:01 +02001686 * Duplicate the contents of ipipe to opipe without actually
1687 * copying the data.
Jens Axboe70524492006-04-11 15:51:17 +02001688 */
Jens Axboeaadd06e2006-07-10 11:00:01 +02001689 if (ipipe && opipe && ipipe != opipe) {
1690 /*
1691 * Keep going, unless we encounter an error. The ipipe/opipe
1692 * ordering doesn't really matter.
1693 */
1694 ret = link_ipipe_prep(ipipe, flags);
1695 if (!ret) {
1696 ret = link_opipe_prep(opipe, flags);
Jens Axboe02cf01a2008-02-20 10:34:51 +01001697 if (!ret)
Jens Axboeaadd06e2006-07-10 11:00:01 +02001698 ret = link_pipe(ipipe, opipe, len, flags);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001699 }
1700 }
Jens Axboe70524492006-04-11 15:51:17 +02001701
Jens Axboeaadd06e2006-07-10 11:00:01 +02001702 return ret;
Jens Axboe70524492006-04-11 15:51:17 +02001703}
1704
1705asmlinkage long sys_tee(int fdin, int fdout, size_t len, unsigned int flags)
1706{
1707 struct file *in;
1708 int error, fput_in;
1709
1710 if (unlikely(!len))
1711 return 0;
1712
1713 error = -EBADF;
1714 in = fget_light(fdin, &fput_in);
1715 if (in) {
1716 if (in->f_mode & FMODE_READ) {
1717 int fput_out;
1718 struct file *out = fget_light(fdout, &fput_out);
1719
1720 if (out) {
1721 if (out->f_mode & FMODE_WRITE)
1722 error = do_tee(in, out, len, flags);
1723 fput_light(out, fput_out);
1724 }
1725 }
1726 fput_light(in, fput_in);
1727 }
1728
1729 return error;
1730}