blob: 0a0b79b01d059a1c71b632de1807100578be8269 [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 Axboecac36bb2007-06-14 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 Axboecac36bb2007-06-14 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 Axboecac36bb2007-06-14 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 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,
Nick Piggin2ae88142006-10-28 10:38:23 -0700323 GFP_KERNEL);
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) {
373 if (TestSetPageLocked(page))
374 break;
375 } else
376 lock_page(page);
Jens Axboe7480a902006-04-11 13:52:47 +0200377
378 /*
379 * page was truncated, stop here. if this isn't the
380 * first page, we'll just complete what we already
381 * added
382 */
383 if (!page->mapping) {
384 unlock_page(page);
Jens Axboe7480a902006-04-11 13:52:47 +0200385 break;
386 }
387 /*
388 * page was already under io and is now done, great
389 */
390 if (PageUptodate(page)) {
391 unlock_page(page);
392 goto fill_it;
393 }
394
Jens Axboe7480a902006-04-11 13:52:47 +0200395 /*
396 * need to read in the page
397 */
398 error = mapping->a_ops->readpage(in, page);
Jens Axboe7480a902006-04-11 13:52:47 +0200399 if (unlikely(error)) {
Jens Axboeeb207962006-04-27 11:05:22 +0200400 /*
401 * We really should re-lookup the page here,
402 * but it complicates things a lot. Instead
403 * lets just do what we already stored, and
404 * we'll get it the next time we are called.
405 */
Jens Axboe7480a902006-04-11 13:52:47 +0200406 if (error == AOP_TRUNCATED_PAGE)
Jens Axboeeb207962006-04-27 11:05:22 +0200407 error = 0;
408
Jens Axboe7480a902006-04-11 13:52:47 +0200409 break;
410 }
Jens Axboe620a3242007-06-07 09:39:42 +0200411 }
412fill_it:
413 /*
414 * i_size must be checked after PageUptodate.
415 */
416 isize = i_size_read(mapping->host);
417 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
418 if (unlikely(!isize || index > end_index))
419 break;
420
421 /*
422 * if this is the last page, see if we need to shrink
423 * the length and stop
424 */
425 if (end_index == index) {
426 unsigned int plen;
Jens Axboe91ad66e2006-04-19 15:55:10 +0200427
428 /*
Jens Axboe620a3242007-06-07 09:39:42 +0200429 * max good bytes in this page
Jens Axboe91ad66e2006-04-19 15:55:10 +0200430 */
Jens Axboe620a3242007-06-07 09:39:42 +0200431 plen = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
432 if (plen <= loff)
Jens Axboe91ad66e2006-04-19 15:55:10 +0200433 break;
Jens Axboe91ad66e2006-04-19 15:55:10 +0200434
435 /*
Jens Axboe620a3242007-06-07 09:39:42 +0200436 * force quit after adding this page
Jens Axboe91ad66e2006-04-19 15:55:10 +0200437 */
Jens Axboe620a3242007-06-07 09:39:42 +0200438 this_len = min(this_len, plen - loff);
439 len = this_len;
Jens Axboe7480a902006-04-11 13:52:47 +0200440 }
Jens Axboe620a3242007-06-07 09:39:42 +0200441
Jens Axboeeb207962006-04-27 11:05:22 +0200442 partial[page_nr].offset = loff;
443 partial[page_nr].len = this_len;
Jens Axboe82aa5d62006-04-20 13:05:48 +0200444 len -= this_len;
Jens Axboe91ad66e2006-04-19 15:55:10 +0200445 loff = 0;
Jens Axboeeb207962006-04-27 11:05:22 +0200446 spd.nr_pages++;
447 index++;
Jens Axboe5274f052006-03-30 15:15:30 +0200448 }
449
Jens Axboeeb207962006-04-27 11:05:22 +0200450 /*
Hugh Dickins475ecad2007-06-07 09:36:00 +0200451 * Release any pages at the end, if we quit early. 'page_nr' is how far
Jens Axboeeb207962006-04-27 11:05:22 +0200452 * we got, 'nr_pages' is how many pages are in the map.
453 */
454 while (page_nr < nr_pages)
455 page_cache_release(pages[page_nr++]);
Fengguang Wuf4e6b492007-10-16 01:24:33 -0700456 in->f_ra.prev_pos = (loff_t)index << PAGE_CACHE_SHIFT;
Jens Axboeeb207962006-04-27 11:05:22 +0200457
Jens Axboe912d35f2006-04-26 10:59:21 +0200458 if (spd.nr_pages)
Jens Axboe00522fb2006-04-26 14:39:29 +0200459 return splice_to_pipe(pipe, &spd);
Jens Axboe5274f052006-03-30 15:15:30 +0200460
Jens Axboe7480a902006-04-11 13:52:47 +0200461 return error;
Jens Axboe5274f052006-03-30 15:15:30 +0200462}
463
Jens Axboe83f91352006-04-02 23:05:09 +0200464/**
465 * generic_file_splice_read - splice data from file to a pipe
466 * @in: file to splice from
Jens Axboe932cc6d2007-06-21 13:10:21 +0200467 * @ppos: position in @in
Jens Axboe83f91352006-04-02 23:05:09 +0200468 * @pipe: pipe to splice to
469 * @len: number of bytes to splice
470 * @flags: splice modifier flags
471 *
Jens Axboe932cc6d2007-06-21 13:10:21 +0200472 * Description:
473 * Will read pages from given file and fill them into a pipe. Can be
474 * used as long as the address_space operations for the source implements
475 * a readpage() hook.
476 *
Jens Axboe83f91352006-04-02 23:05:09 +0200477 */
Jens Axboecbb7e572006-04-11 14:57:50 +0200478ssize_t generic_file_splice_read(struct file *in, loff_t *ppos,
479 struct pipe_inode_info *pipe, size_t len,
480 unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +0200481{
482 ssize_t spliced;
483 int ret;
Jens Axboed366d3982007-06-01 14:54:11 +0200484 loff_t isize, left;
485
486 isize = i_size_read(in->f_mapping->host);
487 if (unlikely(*ppos >= isize))
488 return 0;
489
490 left = isize - *ppos;
491 if (unlikely(left < len))
492 len = left;
Jens Axboe5274f052006-03-30 15:15:30 +0200493
494 ret = 0;
495 spliced = 0;
Jens Axboe51a92c02007-07-13 14:11:43 +0200496 while (len && !spliced) {
Jens Axboecbb7e572006-04-11 14:57:50 +0200497 ret = __generic_file_splice_read(in, ppos, pipe, len, flags);
Jens Axboe5274f052006-03-30 15:15:30 +0200498
Jens Axboec4f895c2006-04-19 15:56:12 +0200499 if (ret < 0)
Jens Axboe5274f052006-03-30 15:15:30 +0200500 break;
Jens Axboec4f895c2006-04-19 15:56:12 +0200501 else if (!ret) {
502 if (spliced)
503 break;
504 if (flags & SPLICE_F_NONBLOCK) {
505 ret = -EAGAIN;
506 break;
507 }
508 }
Jens Axboe5274f052006-03-30 15:15:30 +0200509
Jens Axboecbb7e572006-04-11 14:57:50 +0200510 *ppos += ret;
Jens Axboe5274f052006-03-30 15:15:30 +0200511 len -= ret;
512 spliced += ret;
513 }
514
515 if (spliced)
516 return spliced;
517
518 return ret;
519}
520
Jens Axboe059a8f32006-04-02 23:06:05 +0200521EXPORT_SYMBOL(generic_file_splice_read);
522
Jens Axboe5274f052006-03-30 15:15:30 +0200523/*
Jens Axboe4f6f0bd2006-04-02 23:04:46 +0200524 * Send 'sd->len' bytes to socket from 'sd->file' at position 'sd->pos'
Jens Axboe016b6612006-04-25 15:42:00 +0200525 * using sendpage(). Return the number of bytes sent.
Jens Axboe5274f052006-03-30 15:15:30 +0200526 */
Jens Axboe76ad4d12006-05-03 10:41:33 +0200527static int pipe_to_sendpage(struct pipe_inode_info *pipe,
Jens Axboe5274f052006-03-30 15:15:30 +0200528 struct pipe_buffer *buf, struct splice_desc *sd)
529{
Jens Axboe6a14b902007-06-14 13:08:55 +0200530 struct file *file = sd->u.file;
Jens Axboe5274f052006-03-30 15:15:30 +0200531 loff_t pos = sd->pos;
Jens Axboef84d7512006-05-01 19:59:03 +0200532 int ret, more;
Jens Axboe5274f052006-03-30 15:15:30 +0200533
Jens Axboecac36bb2007-06-14 13:10:48 +0200534 ret = buf->ops->confirm(pipe, buf);
Jens Axboef84d7512006-05-01 19:59:03 +0200535 if (!ret) {
536 more = (sd->flags & SPLICE_F_MORE) || sd->len < sd->total_len;
Jens Axboe5274f052006-03-30 15:15:30 +0200537
Jens Axboef84d7512006-05-01 19:59:03 +0200538 ret = file->f_op->sendpage(file, buf->page, buf->offset,
539 sd->len, &pos, more);
540 }
Jens Axboe5274f052006-03-30 15:15:30 +0200541
Jens Axboe016b6612006-04-25 15:42:00 +0200542 return ret;
Jens Axboe5274f052006-03-30 15:15:30 +0200543}
544
545/*
546 * This is a little more tricky than the file -> pipe splicing. There are
547 * basically three cases:
548 *
549 * - Destination page already exists in the address space and there
550 * are users of it. For that case we have no other option that
551 * copying the data. Tough luck.
552 * - Destination page already exists in the address space, but there
553 * are no users of it. Make sure it's uptodate, then drop it. Fall
554 * through to last case.
555 * - Destination page does not exist, we can add the pipe page to
556 * the page cache and avoid the copy.
557 *
Jens Axboe83f91352006-04-02 23:05:09 +0200558 * If asked to move pages to the output file (SPLICE_F_MOVE is set in
559 * sd->flags), we attempt to migrate pages from the pipe to the output
560 * file address space page cache. This is possible if no one else has
561 * the pipe page referenced outside of the pipe and page cache. If
562 * SPLICE_F_MOVE isn't set, or we cannot move the page, we simply create
563 * a new page in the output file page cache and fill/dirty that.
Jens Axboe5274f052006-03-30 15:15:30 +0200564 */
Jens Axboe76ad4d12006-05-03 10:41:33 +0200565static int pipe_to_file(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
Jens Axboe5274f052006-03-30 15:15:30 +0200566 struct splice_desc *sd)
567{
Jens Axboe6a14b902007-06-14 13:08:55 +0200568 struct file *file = sd->u.file;
Jens Axboe5274f052006-03-30 15:15:30 +0200569 struct address_space *mapping = file->f_mapping;
Jens Axboe016b6612006-04-25 15:42:00 +0200570 unsigned int offset, this_len;
Jens Axboe5274f052006-03-30 15:15:30 +0200571 struct page *page;
Nick Pigginafddba42007-10-16 01:25:01 -0700572 void *fsdata;
Jens Axboe3e7ee3e2006-04-02 23:11:04 +0200573 int ret;
Jens Axboe5274f052006-03-30 15:15:30 +0200574
575 /*
Jens Axboe49d0b212006-04-10 09:04:41 +0200576 * make sure the data in this buffer is uptodate
Jens Axboe5274f052006-03-30 15:15:30 +0200577 */
Jens Axboecac36bb2007-06-14 13:10:48 +0200578 ret = buf->ops->confirm(pipe, buf);
Jens Axboef84d7512006-05-01 19:59:03 +0200579 if (unlikely(ret))
580 return ret;
Jens Axboe5274f052006-03-30 15:15:30 +0200581
Jens Axboe5274f052006-03-30 15:15:30 +0200582 offset = sd->pos & ~PAGE_CACHE_MASK;
583
Jens Axboe016b6612006-04-25 15:42:00 +0200584 this_len = sd->len;
585 if (this_len + offset > PAGE_CACHE_SIZE)
586 this_len = PAGE_CACHE_SIZE - offset;
587
Nick Pigginafddba42007-10-16 01:25:01 -0700588 ret = pagecache_write_begin(file, mapping, sd->pos, this_len,
589 AOP_FLAG_UNINTERRUPTIBLE, &page, &fsdata);
590 if (unlikely(ret))
591 goto out;
Jens Axboe5274f052006-03-30 15:15:30 +0200592
Jens Axboe0568b402006-05-01 19:50:48 +0200593 if (buf->page != page) {
Jens Axboef84d7512006-05-01 19:59:03 +0200594 /*
595 * Careful, ->map() uses KM_USER0!
596 */
Jens Axboe76ad4d12006-05-03 10:41:33 +0200597 char *src = buf->ops->map(pipe, buf, 1);
Jens Axboef84d7512006-05-01 19:59:03 +0200598 char *dst = kmap_atomic(page, KM_USER1);
Jens Axboe5abc97a2006-03-30 15:16:46 +0200599
Jens Axboe016b6612006-04-25 15:42:00 +0200600 memcpy(dst + offset, src + buf->offset, this_len);
Jens Axboe5abc97a2006-03-30 15:16:46 +0200601 flush_dcache_page(page);
Jens Axboef84d7512006-05-01 19:59:03 +0200602 kunmap_atomic(dst, KM_USER1);
Jens Axboe76ad4d12006-05-03 10:41:33 +0200603 buf->ops->unmap(pipe, buf, src);
Jens Axboe5abc97a2006-03-30 15:16:46 +0200604 }
Nick Pigginafddba42007-10-16 01:25:01 -0700605 ret = pagecache_write_end(file, mapping, sd->pos, this_len, this_len,
606 page, fsdata);
Jens Axboe5274f052006-03-30 15:15:30 +0200607out:
Jens Axboe5274f052006-03-30 15:15:30 +0200608 return ret;
609}
610
Jens Axboe932cc6d2007-06-21 13:10:21 +0200611/**
612 * __splice_from_pipe - splice data from a pipe to given actor
613 * @pipe: pipe to splice from
614 * @sd: information to @actor
615 * @actor: handler that splices the data
616 *
617 * Description:
618 * This function does little more than loop over the pipe and call
619 * @actor to do the actual moving of a single struct pipe_buffer to
620 * the desired destination. See pipe_to_file, pipe_to_sendpage, or
621 * pipe_to_user.
622 *
Jens Axboe83f91352006-04-02 23:05:09 +0200623 */
Jens Axboec66ab6f2007-06-12 21:17:17 +0200624ssize_t __splice_from_pipe(struct pipe_inode_info *pipe, struct splice_desc *sd,
625 splice_actor *actor)
Jens Axboe5274f052006-03-30 15:15:30 +0200626{
Jens Axboe5274f052006-03-30 15:15:30 +0200627 int ret, do_wakeup, err;
Jens Axboe5274f052006-03-30 15:15:30 +0200628
629 ret = 0;
630 do_wakeup = 0;
631
Jens Axboe5274f052006-03-30 15:15:30 +0200632 for (;;) {
Jens Axboe6f767b02006-04-11 13:53:56 +0200633 if (pipe->nrbufs) {
634 struct pipe_buffer *buf = pipe->bufs + pipe->curbuf;
Eric Dumazetd4c3cca2006-12-13 00:34:04 -0800635 const struct pipe_buf_operations *ops = buf->ops;
Jens Axboe5274f052006-03-30 15:15:30 +0200636
Jens Axboec66ab6f2007-06-12 21:17:17 +0200637 sd->len = buf->len;
638 if (sd->len > sd->total_len)
639 sd->len = sd->total_len;
Jens Axboe5274f052006-03-30 15:15:30 +0200640
Jens Axboec66ab6f2007-06-12 21:17:17 +0200641 err = actor(pipe, buf, sd);
Jens Axboe016b6612006-04-25 15:42:00 +0200642 if (err <= 0) {
Jens Axboe5274f052006-03-30 15:15:30 +0200643 if (!ret && err != -ENODATA)
644 ret = err;
645
646 break;
647 }
648
Jens Axboe016b6612006-04-25 15:42:00 +0200649 ret += err;
650 buf->offset += err;
651 buf->len -= err;
652
Jens Axboec66ab6f2007-06-12 21:17:17 +0200653 sd->len -= err;
654 sd->pos += err;
655 sd->total_len -= err;
656 if (sd->len)
Jens Axboe016b6612006-04-25 15:42:00 +0200657 continue;
Ingo Molnar73d62d82006-04-11 13:57:21 +0200658
Jens Axboe5274f052006-03-30 15:15:30 +0200659 if (!buf->len) {
660 buf->ops = NULL;
Ingo Molnar3a326a22006-04-10 15:18:35 +0200661 ops->release(pipe, buf);
Jens Axboe6f767b02006-04-11 13:53:56 +0200662 pipe->curbuf = (pipe->curbuf + 1) & (PIPE_BUFFERS - 1);
663 pipe->nrbufs--;
664 if (pipe->inode)
665 do_wakeup = 1;
Jens Axboe5274f052006-03-30 15:15:30 +0200666 }
667
Jens Axboec66ab6f2007-06-12 21:17:17 +0200668 if (!sd->total_len)
Jens Axboe5274f052006-03-30 15:15:30 +0200669 break;
670 }
671
Jens Axboe6f767b02006-04-11 13:53:56 +0200672 if (pipe->nrbufs)
Jens Axboe5274f052006-03-30 15:15:30 +0200673 continue;
Ingo Molnar3a326a22006-04-10 15:18:35 +0200674 if (!pipe->writers)
Jens Axboe5274f052006-03-30 15:15:30 +0200675 break;
Ingo Molnar3a326a22006-04-10 15:18:35 +0200676 if (!pipe->waiting_writers) {
Jens Axboe5274f052006-03-30 15:15:30 +0200677 if (ret)
678 break;
679 }
680
Jens Axboec66ab6f2007-06-12 21:17:17 +0200681 if (sd->flags & SPLICE_F_NONBLOCK) {
Linus Torvalds29e35092006-04-02 12:46:35 -0700682 if (!ret)
683 ret = -EAGAIN;
684 break;
685 }
686
Jens Axboe5274f052006-03-30 15:15:30 +0200687 if (signal_pending(current)) {
688 if (!ret)
689 ret = -ERESTARTSYS;
690 break;
691 }
692
693 if (do_wakeup) {
Jens Axboec0bd1f62006-04-10 09:03:32 +0200694 smp_mb();
Ingo Molnar3a326a22006-04-10 15:18:35 +0200695 if (waitqueue_active(&pipe->wait))
696 wake_up_interruptible_sync(&pipe->wait);
697 kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
Jens Axboe5274f052006-03-30 15:15:30 +0200698 do_wakeup = 0;
699 }
700
Ingo Molnar3a326a22006-04-10 15:18:35 +0200701 pipe_wait(pipe);
Jens Axboe5274f052006-03-30 15:15:30 +0200702 }
703
Jens Axboe5274f052006-03-30 15:15:30 +0200704 if (do_wakeup) {
Jens Axboec0bd1f62006-04-10 09:03:32 +0200705 smp_mb();
Ingo Molnar3a326a22006-04-10 15:18:35 +0200706 if (waitqueue_active(&pipe->wait))
707 wake_up_interruptible(&pipe->wait);
708 kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
Jens Axboe5274f052006-03-30 15:15:30 +0200709 }
710
Jens Axboe5274f052006-03-30 15:15:30 +0200711 return ret;
Jens Axboe5274f052006-03-30 15:15:30 +0200712}
Mark Fasheh40bee442007-03-21 13:11:02 +0100713EXPORT_SYMBOL(__splice_from_pipe);
Jens Axboe5274f052006-03-30 15:15:30 +0200714
Jens Axboe932cc6d2007-06-21 13:10:21 +0200715/**
716 * splice_from_pipe - splice data from a pipe to a file
717 * @pipe: pipe to splice from
718 * @out: file to splice to
719 * @ppos: position in @out
720 * @len: how many bytes to splice
721 * @flags: splice modifier flags
722 * @actor: handler that splices the data
723 *
724 * Description:
725 * See __splice_from_pipe. This function locks the input and output inodes,
726 * otherwise it's identical to __splice_from_pipe().
727 *
728 */
Mark Fasheh6da61802006-10-17 18:43:07 +0200729ssize_t splice_from_pipe(struct pipe_inode_info *pipe, struct file *out,
730 loff_t *ppos, size_t len, unsigned int flags,
731 splice_actor *actor)
732{
733 ssize_t ret;
734 struct inode *inode = out->f_mapping->host;
Jens Axboec66ab6f2007-06-12 21:17:17 +0200735 struct splice_desc sd = {
736 .total_len = len,
737 .flags = flags,
738 .pos = *ppos,
Jens Axboe6a14b902007-06-14 13:08:55 +0200739 .u.file = out,
Jens Axboec66ab6f2007-06-12 21:17:17 +0200740 };
Mark Fasheh6da61802006-10-17 18:43:07 +0200741
742 /*
743 * The actor worker might be calling ->prepare_write and
744 * ->commit_write. Most of the time, these expect i_mutex to
745 * be held. Since this may result in an ABBA deadlock with
746 * pipe->inode, we have to order lock acquiry here.
747 */
748 inode_double_lock(inode, pipe->inode);
Jens Axboec66ab6f2007-06-12 21:17:17 +0200749 ret = __splice_from_pipe(pipe, &sd, actor);
Mark Fasheh6da61802006-10-17 18:43:07 +0200750 inode_double_unlock(inode, pipe->inode);
751
752 return ret;
753}
754
755/**
756 * generic_file_splice_write_nolock - generic_file_splice_write without mutexes
757 * @pipe: pipe info
758 * @out: file to write to
Jens Axboe932cc6d2007-06-21 13:10:21 +0200759 * @ppos: position in @out
Mark Fasheh6da61802006-10-17 18:43:07 +0200760 * @len: number of bytes to splice
761 * @flags: splice modifier flags
762 *
Jens Axboe932cc6d2007-06-21 13:10:21 +0200763 * Description:
764 * Will either move or copy pages (determined by @flags options) from
765 * the given pipe inode to the given file. The caller is responsible
766 * for acquiring i_mutex on both inodes.
Mark Fasheh6da61802006-10-17 18:43:07 +0200767 *
768 */
769ssize_t
770generic_file_splice_write_nolock(struct pipe_inode_info *pipe, struct file *out,
771 loff_t *ppos, size_t len, unsigned int flags)
772{
773 struct address_space *mapping = out->f_mapping;
774 struct inode *inode = mapping->host;
Jens Axboec66ab6f2007-06-12 21:17:17 +0200775 struct splice_desc sd = {
776 .total_len = len,
777 .flags = flags,
778 .pos = *ppos,
Jens Axboe6a14b902007-06-14 13:08:55 +0200779 .u.file = out,
Jens Axboec66ab6f2007-06-12 21:17:17 +0200780 };
Mark Fasheh6da61802006-10-17 18:43:07 +0200781 ssize_t ret;
782 int err;
783
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800784 err = remove_suid(out->f_path.dentry);
Jens Axboe8c34e2d2006-10-17 19:43:22 +0200785 if (unlikely(err))
786 return err;
787
Jens Axboec66ab6f2007-06-12 21:17:17 +0200788 ret = __splice_from_pipe(pipe, &sd, pipe_to_file);
Mark Fasheh6da61802006-10-17 18:43:07 +0200789 if (ret > 0) {
Jens Axboe17ee4f42007-06-15 13:10:37 +0200790 unsigned long nr_pages;
791
Mark Fasheh6da61802006-10-17 18:43:07 +0200792 *ppos += ret;
Jens Axboe17ee4f42007-06-15 13:10:37 +0200793 nr_pages = (ret + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
Mark Fasheh6da61802006-10-17 18:43:07 +0200794
795 /*
796 * If file or inode is SYNC and we actually wrote some data,
797 * sync it.
798 */
799 if (unlikely((out->f_flags & O_SYNC) || IS_SYNC(inode))) {
800 err = generic_osync_inode(inode, mapping,
801 OSYNC_METADATA|OSYNC_DATA);
802
803 if (err)
804 ret = err;
805 }
Jens Axboe17ee4f42007-06-15 13:10:37 +0200806 balance_dirty_pages_ratelimited_nr(mapping, nr_pages);
Mark Fasheh6da61802006-10-17 18:43:07 +0200807 }
808
809 return ret;
810}
811
812EXPORT_SYMBOL(generic_file_splice_write_nolock);
813
Jens Axboe83f91352006-04-02 23:05:09 +0200814/**
815 * generic_file_splice_write - splice data from a pipe to a file
Ingo Molnar3a326a22006-04-10 15:18:35 +0200816 * @pipe: pipe info
Jens Axboe83f91352006-04-02 23:05:09 +0200817 * @out: file to write to
Jens Axboe932cc6d2007-06-21 13:10:21 +0200818 * @ppos: position in @out
Jens Axboe83f91352006-04-02 23:05:09 +0200819 * @len: number of bytes to splice
820 * @flags: splice modifier flags
821 *
Jens Axboe932cc6d2007-06-21 13:10:21 +0200822 * Description:
823 * Will either move or copy pages (determined by @flags options) from
824 * the given pipe inode to the given file.
Jens Axboe83f91352006-04-02 23:05:09 +0200825 *
826 */
Ingo Molnar3a326a22006-04-10 15:18:35 +0200827ssize_t
828generic_file_splice_write(struct pipe_inode_info *pipe, struct file *out,
Jens Axboecbb7e572006-04-11 14:57:50 +0200829 loff_t *ppos, size_t len, unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +0200830{
Jens Axboe4f6f0bd2006-04-02 23:04:46 +0200831 struct address_space *mapping = out->f_mapping;
Jens Axboe8c34e2d2006-10-17 19:43:22 +0200832 struct inode *inode = mapping->host;
Serge E. Hallynb5376772007-10-16 23:31:36 -0700833 int killsuid, killpriv;
Ingo Molnar3a326a22006-04-10 15:18:35 +0200834 ssize_t ret;
Serge E. Hallynb5376772007-10-16 23:31:36 -0700835 int err = 0;
Jens Axboe8c34e2d2006-10-17 19:43:22 +0200836
Serge E. Hallynb5376772007-10-16 23:31:36 -0700837 killpriv = security_inode_need_killpriv(out->f_path.dentry);
838 killsuid = should_remove_suid(out->f_path.dentry);
839 if (unlikely(killsuid || killpriv)) {
Jens Axboe8c34e2d2006-10-17 19:43:22 +0200840 mutex_lock(&inode->i_mutex);
Serge E. Hallynb5376772007-10-16 23:31:36 -0700841 if (killpriv)
842 err = security_inode_killpriv(out->f_path.dentry);
843 if (!err && killsuid)
844 err = __remove_suid(out->f_path.dentry, killsuid);
Jens Axboe8c34e2d2006-10-17 19:43:22 +0200845 mutex_unlock(&inode->i_mutex);
846 if (err)
847 return err;
848 }
Ingo Molnar3a326a22006-04-10 15:18:35 +0200849
Jens Axboe00522fb2006-04-26 14:39:29 +0200850 ret = splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_file);
Jens Axboea4514eb2006-04-19 15:57:05 +0200851 if (ret > 0) {
Jens Axboe17ee4f42007-06-15 13:10:37 +0200852 unsigned long nr_pages;
853
Jens Axboea4514eb2006-04-19 15:57:05 +0200854 *ppos += ret;
Jens Axboe17ee4f42007-06-15 13:10:37 +0200855 nr_pages = (ret + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
Jens Axboe4f6f0bd2006-04-02 23:04:46 +0200856
Jens Axboea4514eb2006-04-19 15:57:05 +0200857 /*
858 * If file or inode is SYNC and we actually wrote some data,
859 * sync it.
860 */
861 if (unlikely((out->f_flags & O_SYNC) || IS_SYNC(inode))) {
Jens Axboea4514eb2006-04-19 15:57:05 +0200862 mutex_lock(&inode->i_mutex);
863 err = generic_osync_inode(inode, mapping,
864 OSYNC_METADATA|OSYNC_DATA);
865 mutex_unlock(&inode->i_mutex);
866
867 if (err)
868 ret = err;
869 }
Jens Axboe17ee4f42007-06-15 13:10:37 +0200870 balance_dirty_pages_ratelimited_nr(mapping, nr_pages);
Jens Axboe4f6f0bd2006-04-02 23:04:46 +0200871 }
872
873 return ret;
Jens Axboe5274f052006-03-30 15:15:30 +0200874}
875
Jens Axboe059a8f32006-04-02 23:06:05 +0200876EXPORT_SYMBOL(generic_file_splice_write);
877
Jens Axboe83f91352006-04-02 23:05:09 +0200878/**
879 * generic_splice_sendpage - splice data from a pipe to a socket
Jens Axboe932cc6d2007-06-21 13:10:21 +0200880 * @pipe: pipe to splice from
Jens Axboe83f91352006-04-02 23:05:09 +0200881 * @out: socket to write to
Jens Axboe932cc6d2007-06-21 13:10:21 +0200882 * @ppos: position in @out
Jens Axboe83f91352006-04-02 23:05:09 +0200883 * @len: number of bytes to splice
884 * @flags: splice modifier flags
885 *
Jens Axboe932cc6d2007-06-21 13:10:21 +0200886 * Description:
887 * Will send @len bytes from the pipe to a network socket. No data copying
888 * is involved.
Jens Axboe83f91352006-04-02 23:05:09 +0200889 *
890 */
Ingo Molnar3a326a22006-04-10 15:18:35 +0200891ssize_t generic_splice_sendpage(struct pipe_inode_info *pipe, struct file *out,
Jens Axboecbb7e572006-04-11 14:57:50 +0200892 loff_t *ppos, size_t len, unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +0200893{
Jens Axboe00522fb2006-04-26 14:39:29 +0200894 return splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_sendpage);
Jens Axboe5274f052006-03-30 15:15:30 +0200895}
896
Jens Axboe059a8f32006-04-02 23:06:05 +0200897EXPORT_SYMBOL(generic_splice_sendpage);
Jeff Garzika0f06782006-03-30 23:06:13 -0500898
Jens Axboe83f91352006-04-02 23:05:09 +0200899/*
900 * Attempt to initiate a splice from pipe to file.
901 */
Ingo Molnar3a326a22006-04-10 15:18:35 +0200902static long do_splice_from(struct pipe_inode_info *pipe, struct file *out,
Jens Axboecbb7e572006-04-11 14:57:50 +0200903 loff_t *ppos, size_t len, 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(!out->f_op || !out->f_op->splice_write))
Jens Axboe5274f052006-03-30 15:15:30 +0200908 return -EINVAL;
909
Jens Axboe49570e92006-04-11 13:56:09 +0200910 if (unlikely(!(out->f_mode & FMODE_WRITE)))
Jens Axboe5274f052006-03-30 15:15:30 +0200911 return -EBADF;
912
Jens Axboecbb7e572006-04-11 14:57:50 +0200913 ret = rw_verify_area(WRITE, out, 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 out->f_op->splice_write(pipe, out, ppos, len, flags);
Jens Axboe5274f052006-03-30 15:15:30 +0200918}
919
Jens Axboe83f91352006-04-02 23:05:09 +0200920/*
921 * Attempt to initiate a splice from a file to a pipe.
922 */
Jens Axboecbb7e572006-04-11 14:57:50 +0200923static long do_splice_to(struct file *in, loff_t *ppos,
924 struct pipe_inode_info *pipe, size_t len,
925 unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +0200926{
Jens Axboe5274f052006-03-30 15:15:30 +0200927 int ret;
928
Jens Axboe49570e92006-04-11 13:56:09 +0200929 if (unlikely(!in->f_op || !in->f_op->splice_read))
Jens Axboe5274f052006-03-30 15:15:30 +0200930 return -EINVAL;
931
Jens Axboe49570e92006-04-11 13:56:09 +0200932 if (unlikely(!(in->f_mode & FMODE_READ)))
Jens Axboe5274f052006-03-30 15:15:30 +0200933 return -EBADF;
934
Jens Axboecbb7e572006-04-11 14:57:50 +0200935 ret = rw_verify_area(READ, in, ppos, len);
Jens Axboe5274f052006-03-30 15:15:30 +0200936 if (unlikely(ret < 0))
937 return ret;
938
Jens Axboecbb7e572006-04-11 14:57:50 +0200939 return in->f_op->splice_read(in, ppos, pipe, len, flags);
Jens Axboe5274f052006-03-30 15:15:30 +0200940}
941
Jens Axboe932cc6d2007-06-21 13:10:21 +0200942/**
943 * splice_direct_to_actor - splices data directly between two non-pipes
944 * @in: file to splice from
945 * @sd: actor information on where to splice to
946 * @actor: handles the data splicing
947 *
948 * Description:
949 * This is a special case helper to splice directly between two
950 * points, without requiring an explicit pipe. Internally an allocated
Randy Dunlap79685b82007-07-27 08:08:51 +0200951 * pipe is cached in the process, and reused during the lifetime of
Jens Axboe932cc6d2007-06-21 13:10:21 +0200952 * that process.
953 *
Jens Axboec66ab6f2007-06-12 21:17:17 +0200954 */
955ssize_t splice_direct_to_actor(struct file *in, struct splice_desc *sd,
956 splice_direct_actor *actor)
Jens Axboeb92ce552006-04-11 13:52:07 +0200957{
958 struct pipe_inode_info *pipe;
959 long ret, bytes;
960 umode_t i_mode;
Jens Axboec66ab6f2007-06-12 21:17:17 +0200961 size_t len;
962 int i, flags;
Jens Axboeb92ce552006-04-11 13:52:07 +0200963
964 /*
965 * We require the input being a regular file, as we don't want to
966 * randomly drop data for eg socket -> socket splicing. Use the
967 * piped splicing for that!
968 */
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800969 i_mode = in->f_path.dentry->d_inode->i_mode;
Jens Axboeb92ce552006-04-11 13:52:07 +0200970 if (unlikely(!S_ISREG(i_mode) && !S_ISBLK(i_mode)))
971 return -EINVAL;
972
973 /*
974 * neither in nor out is a pipe, setup an internal pipe attached to
975 * 'out' and transfer the wanted data from 'in' to 'out' through that
976 */
977 pipe = current->splice_pipe;
Jens Axboe49570e92006-04-11 13:56:09 +0200978 if (unlikely(!pipe)) {
Jens Axboeb92ce552006-04-11 13:52:07 +0200979 pipe = alloc_pipe_info(NULL);
980 if (!pipe)
981 return -ENOMEM;
982
983 /*
984 * We don't have an immediate reader, but we'll read the stuff
Jens Axboe00522fb2006-04-26 14:39:29 +0200985 * out of the pipe right after the splice_to_pipe(). So set
Jens Axboeb92ce552006-04-11 13:52:07 +0200986 * PIPE_READERS appropriately.
987 */
988 pipe->readers = 1;
989
990 current->splice_pipe = pipe;
991 }
992
993 /*
Ingo Molnar73d62d82006-04-11 13:57:21 +0200994 * Do the splice.
Jens Axboeb92ce552006-04-11 13:52:07 +0200995 */
996 ret = 0;
997 bytes = 0;
Jens Axboec66ab6f2007-06-12 21:17:17 +0200998 len = sd->total_len;
999 flags = sd->flags;
1000
1001 /*
1002 * Don't block on output, we have to drain the direct pipe.
1003 */
1004 sd->flags &= ~SPLICE_F_NONBLOCK;
Jens Axboeb92ce552006-04-11 13:52:07 +02001005
1006 while (len) {
Jens Axboe51a92c02007-07-13 14:11:43 +02001007 size_t read_len;
Jens Axboebcd4f3a2007-07-16 14:41:49 +02001008 loff_t pos = sd->pos;
Jens Axboeb92ce552006-04-11 13:52:07 +02001009
Jens Axboebcd4f3a2007-07-16 14:41:49 +02001010 ret = do_splice_to(in, &pos, pipe, len, flags);
Jens Axboe51a92c02007-07-13 14:11:43 +02001011 if (unlikely(ret <= 0))
Jens Axboeb92ce552006-04-11 13:52:07 +02001012 goto out_release;
1013
1014 read_len = ret;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001015 sd->total_len = read_len;
Jens Axboeb92ce552006-04-11 13:52:07 +02001016
1017 /*
1018 * NOTE: nonblocking mode only applies to the input. We
1019 * must not do the output in nonblocking mode as then we
1020 * could get stuck data in the internal pipe:
1021 */
Jens Axboec66ab6f2007-06-12 21:17:17 +02001022 ret = actor(pipe, sd);
Jens Axboe51a92c02007-07-13 14:11:43 +02001023 if (unlikely(ret <= 0))
Jens Axboeb92ce552006-04-11 13:52:07 +02001024 goto out_release;
1025
1026 bytes += ret;
1027 len -= ret;
Jens Axboebcd4f3a2007-07-16 14:41:49 +02001028 sd->pos = pos;
Jens Axboeb92ce552006-04-11 13:52:07 +02001029
Jens Axboe51a92c02007-07-13 14:11:43 +02001030 if (ret < read_len)
1031 goto out_release;
Jens Axboeb92ce552006-04-11 13:52:07 +02001032 }
1033
1034 pipe->nrbufs = pipe->curbuf = 0;
Jens Axboeb92ce552006-04-11 13:52:07 +02001035 return bytes;
1036
1037out_release:
1038 /*
1039 * If we did an incomplete transfer we must release
1040 * the pipe buffers in question:
1041 */
1042 for (i = 0; i < PIPE_BUFFERS; i++) {
1043 struct pipe_buffer *buf = pipe->bufs + i;
1044
1045 if (buf->ops) {
1046 buf->ops->release(pipe, buf);
1047 buf->ops = NULL;
1048 }
1049 }
1050 pipe->nrbufs = pipe->curbuf = 0;
1051
1052 /*
1053 * If we transferred some data, return the number of bytes:
1054 */
1055 if (bytes > 0)
1056 return bytes;
1057
1058 return ret;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001059
1060}
1061EXPORT_SYMBOL(splice_direct_to_actor);
1062
1063static int direct_splice_actor(struct pipe_inode_info *pipe,
1064 struct splice_desc *sd)
1065{
Jens Axboe6a14b902007-06-14 13:08:55 +02001066 struct file *file = sd->u.file;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001067
1068 return do_splice_from(pipe, file, &sd->pos, sd->total_len, sd->flags);
1069}
1070
Jens Axboe932cc6d2007-06-21 13:10:21 +02001071/**
1072 * do_splice_direct - splices data directly between two files
1073 * @in: file to splice from
1074 * @ppos: input file offset
1075 * @out: file to splice to
1076 * @len: number of bytes to splice
1077 * @flags: splice modifier flags
1078 *
1079 * Description:
1080 * For use by do_sendfile(). splice can easily emulate sendfile, but
1081 * doing it in the application would incur an extra system call
1082 * (splice in + splice out, as compared to just sendfile()). So this helper
1083 * can splice directly through a process-private pipe.
1084 *
1085 */
Jens Axboec66ab6f2007-06-12 21:17:17 +02001086long do_splice_direct(struct file *in, loff_t *ppos, struct file *out,
1087 size_t len, unsigned int flags)
1088{
1089 struct splice_desc sd = {
1090 .len = len,
1091 .total_len = len,
1092 .flags = flags,
1093 .pos = *ppos,
Jens Axboe6a14b902007-06-14 13:08:55 +02001094 .u.file = out,
Jens Axboec66ab6f2007-06-12 21:17:17 +02001095 };
Jens Axboe51a92c02007-07-13 14:11:43 +02001096 long ret;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001097
1098 ret = splice_direct_to_actor(in, &sd, direct_splice_actor);
Jens Axboe51a92c02007-07-13 14:11:43 +02001099 if (ret > 0)
1100 *ppos += ret;
1101
Jens Axboec66ab6f2007-06-12 21:17:17 +02001102 return ret;
Jens Axboeb92ce552006-04-11 13:52:07 +02001103}
1104
Jens Axboe83f91352006-04-02 23:05:09 +02001105/*
Jens Axboeddac0d32006-11-04 12:49:32 +01001106 * After the inode slimming patch, i_pipe/i_bdev/i_cdev share the same
1107 * location, so checking ->i_pipe is not enough to verify that this is a
1108 * pipe.
1109 */
1110static inline struct pipe_inode_info *pipe_info(struct inode *inode)
1111{
1112 if (S_ISFIFO(inode->i_mode))
1113 return inode->i_pipe;
1114
1115 return NULL;
1116}
1117
1118/*
Jens Axboe83f91352006-04-02 23:05:09 +02001119 * Determine where to splice to/from.
1120 */
Ingo Molnar529565d2006-04-10 15:18:58 +02001121static long do_splice(struct file *in, loff_t __user *off_in,
1122 struct file *out, loff_t __user *off_out,
1123 size_t len, unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +02001124{
Ingo Molnar3a326a22006-04-10 15:18:35 +02001125 struct pipe_inode_info *pipe;
Jens Axboecbb7e572006-04-11 14:57:50 +02001126 loff_t offset, *off;
Jens Axboea4514eb2006-04-19 15:57:05 +02001127 long ret;
Jens Axboe5274f052006-03-30 15:15:30 +02001128
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001129 pipe = pipe_info(in->f_path.dentry->d_inode);
Ingo Molnar529565d2006-04-10 15:18:58 +02001130 if (pipe) {
1131 if (off_in)
1132 return -ESPIPE;
Jens Axboeb92ce552006-04-11 13:52:07 +02001133 if (off_out) {
1134 if (out->f_op->llseek == no_llseek)
1135 return -EINVAL;
Jens Axboecbb7e572006-04-11 14:57:50 +02001136 if (copy_from_user(&offset, off_out, sizeof(loff_t)))
Jens Axboeb92ce552006-04-11 13:52:07 +02001137 return -EFAULT;
Jens Axboecbb7e572006-04-11 14:57:50 +02001138 off = &offset;
1139 } else
1140 off = &out->f_pos;
Ingo Molnar529565d2006-04-10 15:18:58 +02001141
Jens Axboea4514eb2006-04-19 15:57:05 +02001142 ret = do_splice_from(pipe, out, off, len, flags);
1143
1144 if (off_out && copy_to_user(off_out, off, sizeof(loff_t)))
1145 ret = -EFAULT;
1146
1147 return ret;
Ingo Molnar529565d2006-04-10 15:18:58 +02001148 }
Jens Axboe5274f052006-03-30 15:15:30 +02001149
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001150 pipe = pipe_info(out->f_path.dentry->d_inode);
Ingo Molnar529565d2006-04-10 15:18:58 +02001151 if (pipe) {
1152 if (off_out)
1153 return -ESPIPE;
Jens Axboeb92ce552006-04-11 13:52:07 +02001154 if (off_in) {
1155 if (in->f_op->llseek == no_llseek)
1156 return -EINVAL;
Jens Axboecbb7e572006-04-11 14:57:50 +02001157 if (copy_from_user(&offset, off_in, sizeof(loff_t)))
Jens Axboeb92ce552006-04-11 13:52:07 +02001158 return -EFAULT;
Jens Axboecbb7e572006-04-11 14:57:50 +02001159 off = &offset;
1160 } else
1161 off = &in->f_pos;
Ingo Molnar529565d2006-04-10 15:18:58 +02001162
Jens Axboea4514eb2006-04-19 15:57:05 +02001163 ret = do_splice_to(in, off, pipe, len, flags);
1164
1165 if (off_in && copy_to_user(off_in, off, sizeof(loff_t)))
1166 ret = -EFAULT;
1167
1168 return ret;
Ingo Molnar529565d2006-04-10 15:18:58 +02001169 }
Jens Axboe5274f052006-03-30 15:15:30 +02001170
1171 return -EINVAL;
1172}
1173
Jens Axboe912d35f2006-04-26 10:59:21 +02001174/*
Linus Torvalds75723952007-10-01 13:17:28 -07001175 * Do a copy-from-user while holding the mmap_semaphore for reading, in a
1176 * manner safe from deadlocking with simultaneous mmap() (grabbing mmap_sem
1177 * for writing) and page faulting on the user memory pointed to by src.
1178 * This assumes that we will very rarely hit the partial != 0 path, or this
1179 * will not be a win.
1180 */
1181static int copy_from_user_mmap_sem(void *dst, const void __user *src, size_t n)
1182{
1183 int partial;
1184
1185 pagefault_disable();
1186 partial = __copy_from_user_inatomic(dst, src, n);
1187 pagefault_enable();
1188
1189 /*
1190 * Didn't copy everything, drop the mmap_sem and do a faulting copy
1191 */
1192 if (unlikely(partial)) {
1193 up_read(&current->mm->mmap_sem);
1194 partial = copy_from_user(dst, src, n);
1195 down_read(&current->mm->mmap_sem);
1196 }
1197
1198 return partial;
1199}
1200
1201/*
Jens Axboe912d35f2006-04-26 10:59:21 +02001202 * Map an iov into an array of pages and offset/length tupples. With the
1203 * partial_page structure, we can map several non-contiguous ranges into
1204 * our ones pages[] map instead of splitting that operation into pieces.
1205 * Could easily be exported as a generic helper for other users, in which
1206 * case one would probably want to add a 'max_nr_pages' parameter as well.
1207 */
1208static int get_iovec_page_array(const struct iovec __user *iov,
1209 unsigned int nr_vecs, struct page **pages,
Jens Axboe7afa6fd2006-05-01 20:02:33 +02001210 struct partial_page *partial, int aligned)
Jens Axboe912d35f2006-04-26 10:59:21 +02001211{
1212 int buffers = 0, error = 0;
1213
Jens Axboe912d35f2006-04-26 10:59:21 +02001214 down_read(&current->mm->mmap_sem);
1215
1216 while (nr_vecs) {
1217 unsigned long off, npages;
Linus Torvalds75723952007-10-01 13:17:28 -07001218 struct iovec entry;
Jens Axboe912d35f2006-04-26 10:59:21 +02001219 void __user *base;
1220 size_t len;
1221 int i;
1222
Linus Torvalds75723952007-10-01 13:17:28 -07001223 error = -EFAULT;
1224 if (copy_from_user_mmap_sem(&entry, iov, sizeof(entry)))
Jens Axboe912d35f2006-04-26 10:59:21 +02001225 break;
Linus Torvalds75723952007-10-01 13:17:28 -07001226
1227 base = entry.iov_base;
1228 len = entry.iov_len;
Jens Axboe912d35f2006-04-26 10:59:21 +02001229
1230 /*
1231 * Sanity check this iovec. 0 read succeeds.
1232 */
Linus Torvalds75723952007-10-01 13:17:28 -07001233 error = 0;
Jens Axboe912d35f2006-04-26 10:59:21 +02001234 if (unlikely(!len))
1235 break;
1236 error = -EFAULT;
1237 if (unlikely(!base))
1238 break;
1239
1240 /*
1241 * Get this base offset and number of pages, then map
1242 * in the user pages.
1243 */
1244 off = (unsigned long) base & ~PAGE_MASK;
Jens Axboe7afa6fd2006-05-01 20:02:33 +02001245
1246 /*
1247 * If asked for alignment, the offset must be zero and the
1248 * length a multiple of the PAGE_SIZE.
1249 */
1250 error = -EINVAL;
1251 if (aligned && (off || len & ~PAGE_MASK))
1252 break;
1253
Jens Axboe912d35f2006-04-26 10:59:21 +02001254 npages = (off + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1255 if (npages > PIPE_BUFFERS - buffers)
1256 npages = PIPE_BUFFERS - buffers;
1257
1258 error = get_user_pages(current, current->mm,
1259 (unsigned long) base, npages, 0, 0,
1260 &pages[buffers], NULL);
1261
1262 if (unlikely(error <= 0))
1263 break;
1264
1265 /*
1266 * Fill this contiguous range into the partial page map.
1267 */
1268 for (i = 0; i < error; i++) {
Jens Axboe75914892006-05-02 12:57:18 +02001269 const int plen = min_t(size_t, len, PAGE_SIZE - off);
Jens Axboe912d35f2006-04-26 10:59:21 +02001270
1271 partial[buffers].offset = off;
1272 partial[buffers].len = plen;
1273
1274 off = 0;
1275 len -= plen;
1276 buffers++;
1277 }
1278
1279 /*
1280 * We didn't complete this iov, stop here since it probably
1281 * means we have to move some of this into a pipe to
1282 * be able to continue.
1283 */
1284 if (len)
1285 break;
1286
1287 /*
1288 * Don't continue if we mapped fewer pages than we asked for,
1289 * or if we mapped the max number of pages that we have
1290 * room for.
1291 */
1292 if (error < npages || buffers == PIPE_BUFFERS)
1293 break;
1294
1295 nr_vecs--;
1296 iov++;
1297 }
1298
1299 up_read(&current->mm->mmap_sem);
1300
1301 if (buffers)
1302 return buffers;
1303
1304 return error;
1305}
1306
Jens Axboe6a14b902007-06-14 13:08:55 +02001307static int pipe_to_user(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
1308 struct splice_desc *sd)
1309{
1310 char *src;
1311 int ret;
1312
Jens Axboecac36bb2007-06-14 13:10:48 +02001313 ret = buf->ops->confirm(pipe, buf);
Jens Axboe6a14b902007-06-14 13:08:55 +02001314 if (unlikely(ret))
1315 return ret;
1316
1317 /*
1318 * See if we can use the atomic maps, by prefaulting in the
1319 * pages and doing an atomic copy
1320 */
1321 if (!fault_in_pages_writeable(sd->u.userptr, sd->len)) {
1322 src = buf->ops->map(pipe, buf, 1);
1323 ret = __copy_to_user_inatomic(sd->u.userptr, src + buf->offset,
1324 sd->len);
1325 buf->ops->unmap(pipe, buf, src);
1326 if (!ret) {
1327 ret = sd->len;
1328 goto out;
1329 }
1330 }
1331
1332 /*
1333 * No dice, use slow non-atomic map and copy
1334 */
1335 src = buf->ops->map(pipe, buf, 0);
1336
1337 ret = sd->len;
1338 if (copy_to_user(sd->u.userptr, src + buf->offset, sd->len))
1339 ret = -EFAULT;
1340
Jens Axboe6866bef2007-10-16 10:01:29 +02001341 buf->ops->unmap(pipe, buf, src);
Jens Axboe6a14b902007-06-14 13:08:55 +02001342out:
1343 if (ret > 0)
1344 sd->u.userptr += ret;
Jens Axboe6a14b902007-06-14 13:08:55 +02001345 return ret;
1346}
1347
1348/*
1349 * For lack of a better implementation, implement vmsplice() to userspace
1350 * as a simple copy of the pipes pages to the user iov.
1351 */
1352static long vmsplice_to_user(struct file *file, const struct iovec __user *iov,
1353 unsigned long nr_segs, unsigned int flags)
1354{
1355 struct pipe_inode_info *pipe;
1356 struct splice_desc sd;
1357 ssize_t size;
1358 int error;
1359 long ret;
1360
1361 pipe = pipe_info(file->f_path.dentry->d_inode);
1362 if (!pipe)
1363 return -EBADF;
1364
1365 if (pipe->inode)
1366 mutex_lock(&pipe->inode->i_mutex);
1367
1368 error = ret = 0;
1369 while (nr_segs) {
1370 void __user *base;
1371 size_t len;
1372
1373 /*
1374 * Get user address base and length for this iovec.
1375 */
1376 error = get_user(base, &iov->iov_base);
1377 if (unlikely(error))
1378 break;
1379 error = get_user(len, &iov->iov_len);
1380 if (unlikely(error))
1381 break;
1382
1383 /*
1384 * Sanity check this iovec. 0 read succeeds.
1385 */
1386 if (unlikely(!len))
1387 break;
1388 if (unlikely(!base)) {
1389 error = -EFAULT;
1390 break;
1391 }
1392
1393 sd.len = 0;
1394 sd.total_len = len;
1395 sd.flags = flags;
1396 sd.u.userptr = base;
1397 sd.pos = 0;
1398
1399 size = __splice_from_pipe(pipe, &sd, pipe_to_user);
1400 if (size < 0) {
1401 if (!ret)
1402 ret = size;
1403
1404 break;
1405 }
1406
1407 ret += size;
1408
1409 if (size < len)
1410 break;
1411
1412 nr_segs--;
1413 iov++;
1414 }
1415
1416 if (pipe->inode)
1417 mutex_unlock(&pipe->inode->i_mutex);
1418
1419 if (!ret)
1420 ret = error;
1421
1422 return ret;
1423}
1424
Jens Axboe912d35f2006-04-26 10:59:21 +02001425/*
1426 * vmsplice splices a user address range into a pipe. It can be thought of
1427 * as splice-from-memory, where the regular splice is splice-from-file (or
1428 * to file). In both cases the output is a pipe, naturally.
Jens Axboe912d35f2006-04-26 10:59:21 +02001429 */
Jens Axboe6a14b902007-06-14 13:08:55 +02001430static long vmsplice_to_pipe(struct file *file, const struct iovec __user *iov,
1431 unsigned long nr_segs, unsigned int flags)
Jens Axboe912d35f2006-04-26 10:59:21 +02001432{
Jens Axboeddac0d32006-11-04 12:49:32 +01001433 struct pipe_inode_info *pipe;
Jens Axboe912d35f2006-04-26 10:59:21 +02001434 struct page *pages[PIPE_BUFFERS];
1435 struct partial_page partial[PIPE_BUFFERS];
1436 struct splice_pipe_desc spd = {
1437 .pages = pages,
1438 .partial = partial,
1439 .flags = flags,
1440 .ops = &user_page_pipe_buf_ops,
Jens Axboebbdfc2f2007-11-06 23:29:47 -08001441 .spd_release = spd_release_page,
Jens Axboe912d35f2006-04-26 10:59:21 +02001442 };
1443
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001444 pipe = pipe_info(file->f_path.dentry->d_inode);
Jens Axboeddac0d32006-11-04 12:49:32 +01001445 if (!pipe)
Jens Axboe912d35f2006-04-26 10:59:21 +02001446 return -EBADF;
Jens Axboe912d35f2006-04-26 10:59:21 +02001447
Jens Axboe7afa6fd2006-05-01 20:02:33 +02001448 spd.nr_pages = get_iovec_page_array(iov, nr_segs, pages, partial,
1449 flags & SPLICE_F_GIFT);
Jens Axboe912d35f2006-04-26 10:59:21 +02001450 if (spd.nr_pages <= 0)
1451 return spd.nr_pages;
1452
Jens Axboe00522fb2006-04-26 14:39:29 +02001453 return splice_to_pipe(pipe, &spd);
Jens Axboe912d35f2006-04-26 10:59:21 +02001454}
1455
Jens Axboe6a14b902007-06-14 13:08:55 +02001456/*
1457 * Note that vmsplice only really supports true splicing _from_ user memory
1458 * to a pipe, not the other way around. Splicing from user memory is a simple
1459 * operation that can be supported without any funky alignment restrictions
1460 * or nasty vm tricks. We simply map in the user memory and fill them into
1461 * a pipe. The reverse isn't quite as easy, though. There are two possible
1462 * solutions for that:
1463 *
1464 * - memcpy() the data internally, at which point we might as well just
1465 * do a regular read() on the buffer anyway.
1466 * - Lots of nasty vm tricks, that are neither fast nor flexible (it
1467 * has restriction limitations on both ends of the pipe).
1468 *
1469 * Currently we punt and implement it as a normal copy, see pipe_to_user().
1470 *
1471 */
Jens Axboe912d35f2006-04-26 10:59:21 +02001472asmlinkage long sys_vmsplice(int fd, const struct iovec __user *iov,
1473 unsigned long nr_segs, unsigned int flags)
1474{
1475 struct file *file;
1476 long error;
1477 int fput;
1478
Jens Axboe6a14b902007-06-14 13:08:55 +02001479 if (unlikely(nr_segs > UIO_MAXIOV))
1480 return -EINVAL;
1481 else if (unlikely(!nr_segs))
1482 return 0;
1483
Jens Axboe912d35f2006-04-26 10:59:21 +02001484 error = -EBADF;
1485 file = fget_light(fd, &fput);
1486 if (file) {
1487 if (file->f_mode & FMODE_WRITE)
Jens Axboe6a14b902007-06-14 13:08:55 +02001488 error = vmsplice_to_pipe(file, iov, nr_segs, flags);
1489 else if (file->f_mode & FMODE_READ)
1490 error = vmsplice_to_user(file, iov, nr_segs, flags);
Jens Axboe912d35f2006-04-26 10:59:21 +02001491
1492 fput_light(file, fput);
1493 }
1494
1495 return error;
1496}
1497
Ingo Molnar529565d2006-04-10 15:18:58 +02001498asmlinkage long sys_splice(int fd_in, loff_t __user *off_in,
1499 int fd_out, loff_t __user *off_out,
1500 size_t len, unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +02001501{
1502 long error;
1503 struct file *in, *out;
1504 int fput_in, fput_out;
1505
1506 if (unlikely(!len))
1507 return 0;
1508
1509 error = -EBADF;
Ingo Molnar529565d2006-04-10 15:18:58 +02001510 in = fget_light(fd_in, &fput_in);
Jens Axboe5274f052006-03-30 15:15:30 +02001511 if (in) {
1512 if (in->f_mode & FMODE_READ) {
Ingo Molnar529565d2006-04-10 15:18:58 +02001513 out = fget_light(fd_out, &fput_out);
Jens Axboe5274f052006-03-30 15:15:30 +02001514 if (out) {
1515 if (out->f_mode & FMODE_WRITE)
Ingo Molnar529565d2006-04-10 15:18:58 +02001516 error = do_splice(in, off_in,
1517 out, off_out,
1518 len, flags);
Jens Axboe5274f052006-03-30 15:15:30 +02001519 fput_light(out, fput_out);
1520 }
1521 }
1522
1523 fput_light(in, fput_in);
1524 }
1525
1526 return error;
1527}
Jens Axboe70524492006-04-11 15:51:17 +02001528
1529/*
Jens Axboeaadd06e2006-07-10 11:00:01 +02001530 * Make sure there's data to read. Wait for input if we can, otherwise
1531 * return an appropriate error.
1532 */
1533static int link_ipipe_prep(struct pipe_inode_info *pipe, unsigned int flags)
1534{
1535 int ret;
1536
1537 /*
1538 * Check ->nrbufs without the inode lock first. This function
1539 * is speculative anyways, so missing one is ok.
1540 */
1541 if (pipe->nrbufs)
1542 return 0;
1543
1544 ret = 0;
1545 mutex_lock(&pipe->inode->i_mutex);
1546
1547 while (!pipe->nrbufs) {
1548 if (signal_pending(current)) {
1549 ret = -ERESTARTSYS;
1550 break;
1551 }
1552 if (!pipe->writers)
1553 break;
1554 if (!pipe->waiting_writers) {
1555 if (flags & SPLICE_F_NONBLOCK) {
1556 ret = -EAGAIN;
1557 break;
1558 }
1559 }
1560 pipe_wait(pipe);
1561 }
1562
1563 mutex_unlock(&pipe->inode->i_mutex);
1564 return ret;
1565}
1566
1567/*
1568 * Make sure there's writeable room. Wait for room if we can, otherwise
1569 * return an appropriate error.
1570 */
1571static int link_opipe_prep(struct pipe_inode_info *pipe, unsigned int flags)
1572{
1573 int ret;
1574
1575 /*
1576 * Check ->nrbufs without the inode lock first. This function
1577 * is speculative anyways, so missing one is ok.
1578 */
1579 if (pipe->nrbufs < PIPE_BUFFERS)
1580 return 0;
1581
1582 ret = 0;
1583 mutex_lock(&pipe->inode->i_mutex);
1584
1585 while (pipe->nrbufs >= PIPE_BUFFERS) {
1586 if (!pipe->readers) {
1587 send_sig(SIGPIPE, current, 0);
1588 ret = -EPIPE;
1589 break;
1590 }
1591 if (flags & SPLICE_F_NONBLOCK) {
1592 ret = -EAGAIN;
1593 break;
1594 }
1595 if (signal_pending(current)) {
1596 ret = -ERESTARTSYS;
1597 break;
1598 }
1599 pipe->waiting_writers++;
1600 pipe_wait(pipe);
1601 pipe->waiting_writers--;
1602 }
1603
1604 mutex_unlock(&pipe->inode->i_mutex);
1605 return ret;
1606}
1607
1608/*
Jens Axboe70524492006-04-11 15:51:17 +02001609 * Link contents of ipipe to opipe.
1610 */
1611static int link_pipe(struct pipe_inode_info *ipipe,
1612 struct pipe_inode_info *opipe,
1613 size_t len, unsigned int flags)
1614{
1615 struct pipe_buffer *ibuf, *obuf;
Jens Axboeaadd06e2006-07-10 11:00:01 +02001616 int ret = 0, i = 0, nbuf;
Jens Axboe70524492006-04-11 15:51:17 +02001617
1618 /*
1619 * Potential ABBA deadlock, work around it by ordering lock
1620 * grabbing by inode address. Otherwise two different processes
1621 * could deadlock (one doing tee from A -> B, the other from B -> A).
1622 */
Mark Fasheh62752ee2006-10-17 10:31:38 +02001623 inode_double_lock(ipipe->inode, opipe->inode);
Jens Axboe70524492006-04-11 15:51:17 +02001624
Jens Axboeaadd06e2006-07-10 11:00:01 +02001625 do {
Jens Axboe70524492006-04-11 15:51:17 +02001626 if (!opipe->readers) {
1627 send_sig(SIGPIPE, current, 0);
1628 if (!ret)
1629 ret = -EPIPE;
1630 break;
1631 }
Jens Axboe70524492006-04-11 15:51:17 +02001632
1633 /*
Jens Axboeaadd06e2006-07-10 11:00:01 +02001634 * If we have iterated all input buffers or ran out of
1635 * output room, break.
Jens Axboe70524492006-04-11 15:51:17 +02001636 */
Jens Axboeaadd06e2006-07-10 11:00:01 +02001637 if (i >= ipipe->nrbufs || opipe->nrbufs >= PIPE_BUFFERS)
Jens Axboe70524492006-04-11 15:51:17 +02001638 break;
Jens Axboeaadd06e2006-07-10 11:00:01 +02001639
1640 ibuf = ipipe->bufs + ((ipipe->curbuf + i) & (PIPE_BUFFERS - 1));
1641 nbuf = (opipe->curbuf + opipe->nrbufs) & (PIPE_BUFFERS - 1);
1642
Jens Axboe2a272502006-04-19 15:56:40 +02001643 /*
Jens Axboeaadd06e2006-07-10 11:00:01 +02001644 * Get a reference to this pipe buffer,
1645 * so we can copy the contents over.
Jens Axboe2a272502006-04-19 15:56:40 +02001646 */
Jens Axboeaadd06e2006-07-10 11:00:01 +02001647 ibuf->ops->get(ipipe, ibuf);
Jens Axboe70524492006-04-11 15:51:17 +02001648
Jens Axboeaadd06e2006-07-10 11:00:01 +02001649 obuf = opipe->bufs + nbuf;
1650 *obuf = *ibuf;
Jens Axboe70524492006-04-11 15:51:17 +02001651
Jens Axboeaadd06e2006-07-10 11:00:01 +02001652 /*
1653 * Don't inherit the gift flag, we need to
1654 * prevent multiple steals of this page.
1655 */
1656 obuf->flags &= ~PIPE_BUF_FLAG_GIFT;
1657
1658 if (obuf->len > len)
1659 obuf->len = len;
1660
1661 opipe->nrbufs++;
1662 ret += obuf->len;
1663 len -= obuf->len;
1664 i++;
1665 } while (len);
Jens Axboe70524492006-04-11 15:51:17 +02001666
Mark Fasheh62752ee2006-10-17 10:31:38 +02001667 inode_double_unlock(ipipe->inode, opipe->inode);
Jens Axboe70524492006-04-11 15:51:17 +02001668
Jens Axboeaadd06e2006-07-10 11:00:01 +02001669 /*
1670 * If we put data in the output pipe, wakeup any potential readers.
1671 */
1672 if (ret > 0) {
Jens Axboe70524492006-04-11 15:51:17 +02001673 smp_mb();
1674 if (waitqueue_active(&opipe->wait))
1675 wake_up_interruptible(&opipe->wait);
1676 kill_fasync(&opipe->fasync_readers, SIGIO, POLL_IN);
1677 }
1678
1679 return ret;
1680}
1681
1682/*
1683 * This is a tee(1) implementation that works on pipes. It doesn't copy
1684 * any data, it simply references the 'in' pages on the 'out' pipe.
1685 * The 'flags' used are the SPLICE_F_* variants, currently the only
1686 * applicable one is SPLICE_F_NONBLOCK.
1687 */
1688static long do_tee(struct file *in, struct file *out, size_t len,
1689 unsigned int flags)
1690{
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001691 struct pipe_inode_info *ipipe = pipe_info(in->f_path.dentry->d_inode);
1692 struct pipe_inode_info *opipe = pipe_info(out->f_path.dentry->d_inode);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001693 int ret = -EINVAL;
Jens Axboe70524492006-04-11 15:51:17 +02001694
1695 /*
Jens Axboeaadd06e2006-07-10 11:00:01 +02001696 * Duplicate the contents of ipipe to opipe without actually
1697 * copying the data.
Jens Axboe70524492006-04-11 15:51:17 +02001698 */
Jens Axboeaadd06e2006-07-10 11:00:01 +02001699 if (ipipe && opipe && ipipe != opipe) {
1700 /*
1701 * Keep going, unless we encounter an error. The ipipe/opipe
1702 * ordering doesn't really matter.
1703 */
1704 ret = link_ipipe_prep(ipipe, flags);
1705 if (!ret) {
1706 ret = link_opipe_prep(opipe, flags);
1707 if (!ret) {
1708 ret = link_pipe(ipipe, opipe, len, flags);
1709 if (!ret && (flags & SPLICE_F_NONBLOCK))
1710 ret = -EAGAIN;
1711 }
1712 }
1713 }
Jens Axboe70524492006-04-11 15:51:17 +02001714
Jens Axboeaadd06e2006-07-10 11:00:01 +02001715 return ret;
Jens Axboe70524492006-04-11 15:51:17 +02001716}
1717
1718asmlinkage long sys_tee(int fdin, int fdout, size_t len, unsigned int flags)
1719{
1720 struct file *in;
1721 int error, fput_in;
1722
1723 if (unlikely(!len))
1724 return 0;
1725
1726 error = -EBADF;
1727 in = fget_light(fdin, &fput_in);
1728 if (in) {
1729 if (in->f_mode & FMODE_READ) {
1730 int fput_out;
1731 struct file *out = fget_light(fdout, &fput_out);
1732
1733 if (out) {
1734 if (out->f_mode & FMODE_WRITE)
1735 error = do_tee(in, out, len, flags);
1736 fput_light(out, fput_out);
1737 }
1738 }
1739 fput_light(in, fput_in);
1740 }
1741
1742 return error;
1743}