blob: a1e701c27156f78201532e8f701d1af2266744f0 [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 Axboeca39d652008-05-20 21:27:41 +020061 if (PagePrivate(page) && !try_to_release_page(page, GFP_KERNEL))
62 goto out_unlock;
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 */
Jens Axboeca39d652008-05-20 21:27:41 +020078out_unlock:
Jens Axboe9e94cd42006-06-20 15:01:12 +020079 unlock_page(page);
80 return 1;
Jens Axboe5abc97a2006-03-30 15:16:46 +020081}
82
Jens Axboe76ad4d12006-05-03 10:41:33 +020083static void page_cache_pipe_buf_release(struct pipe_inode_info *pipe,
Jens Axboe5274f052006-03-30 15:15:30 +020084 struct pipe_buffer *buf)
85{
86 page_cache_release(buf->page);
Jens Axboe14328732006-05-03 10:35:26 +020087 buf->flags &= ~PIPE_BUF_FLAG_LRU;
Jens Axboe5274f052006-03-30 15:15:30 +020088}
89
Jens Axboe08457182007-06-12 20:51:32 +020090/*
91 * Check whether the contents of buf is OK to access. Since the content
92 * is a page cache page, IO may be in flight.
93 */
Jens Axboecac36bb02007-06-14 13:10:48 +020094static int page_cache_pipe_buf_confirm(struct pipe_inode_info *pipe,
95 struct pipe_buffer *buf)
Jens Axboe5274f052006-03-30 15:15:30 +020096{
97 struct page *page = buf->page;
Jens Axboe49d0b212006-04-10 09:04:41 +020098 int err;
Jens Axboe5274f052006-03-30 15:15:30 +020099
100 if (!PageUptodate(page)) {
Jens Axboe49d0b212006-04-10 09:04:41 +0200101 lock_page(page);
102
103 /*
104 * Page got truncated/unhashed. This will cause a 0-byte
Ingo Molnar73d62d82006-04-11 13:57:21 +0200105 * splice, if this is the first page.
Jens Axboe49d0b212006-04-10 09:04:41 +0200106 */
107 if (!page->mapping) {
108 err = -ENODATA;
109 goto error;
110 }
111
112 /*
Ingo Molnar73d62d82006-04-11 13:57:21 +0200113 * Uh oh, read-error from disk.
Jens Axboe49d0b212006-04-10 09:04:41 +0200114 */
115 if (!PageUptodate(page)) {
116 err = -EIO;
117 goto error;
118 }
119
120 /*
Jens Axboef84d7512006-05-01 19:59:03 +0200121 * Page is ok afterall, we are done.
Jens Axboe49d0b212006-04-10 09:04:41 +0200122 */
Jens Axboe5274f052006-03-30 15:15:30 +0200123 unlock_page(page);
Jens Axboe5274f052006-03-30 15:15:30 +0200124 }
125
Jens Axboef84d7512006-05-01 19:59:03 +0200126 return 0;
Jens Axboe49d0b212006-04-10 09:04:41 +0200127error:
128 unlock_page(page);
Jens Axboef84d7512006-05-01 19:59:03 +0200129 return err;
Jens Axboe70524492006-04-11 15:51:17 +0200130}
131
Eric Dumazetd4c3cca2006-12-13 00:34:04 -0800132static const struct pipe_buf_operations page_cache_pipe_buf_ops = {
Jens Axboe5274f052006-03-30 15:15:30 +0200133 .can_merge = 0,
Jens Axboef84d7512006-05-01 19:59:03 +0200134 .map = generic_pipe_buf_map,
135 .unmap = generic_pipe_buf_unmap,
Jens Axboecac36bb02007-06-14 13:10:48 +0200136 .confirm = page_cache_pipe_buf_confirm,
Jens Axboe5274f052006-03-30 15:15:30 +0200137 .release = page_cache_pipe_buf_release,
Jens Axboe5abc97a2006-03-30 15:16:46 +0200138 .steal = page_cache_pipe_buf_steal,
Jens Axboef84d7512006-05-01 19:59:03 +0200139 .get = generic_pipe_buf_get,
Jens Axboe5274f052006-03-30 15:15:30 +0200140};
141
Jens Axboe912d35f2006-04-26 10:59:21 +0200142static int user_page_pipe_buf_steal(struct pipe_inode_info *pipe,
143 struct pipe_buffer *buf)
144{
Jens Axboe7afa6fd2006-05-01 20:02:33 +0200145 if (!(buf->flags & PIPE_BUF_FLAG_GIFT))
146 return 1;
147
Jens Axboe14328732006-05-03 10:35:26 +0200148 buf->flags |= PIPE_BUF_FLAG_LRU;
Jens Axboe330ab712006-05-02 15:29:57 +0200149 return generic_pipe_buf_steal(pipe, buf);
Jens Axboe912d35f2006-04-26 10:59:21 +0200150}
151
Eric Dumazetd4c3cca2006-12-13 00:34:04 -0800152static const struct pipe_buf_operations user_page_pipe_buf_ops = {
Jens Axboe912d35f2006-04-26 10:59:21 +0200153 .can_merge = 0,
Jens Axboef84d7512006-05-01 19:59:03 +0200154 .map = generic_pipe_buf_map,
155 .unmap = generic_pipe_buf_unmap,
Jens Axboecac36bb02007-06-14 13:10:48 +0200156 .confirm = generic_pipe_buf_confirm,
Jens Axboe912d35f2006-04-26 10:59:21 +0200157 .release = page_cache_pipe_buf_release,
158 .steal = user_page_pipe_buf_steal,
Jens Axboef84d7512006-05-01 19:59:03 +0200159 .get = generic_pipe_buf_get,
Jens Axboe912d35f2006-04-26 10:59:21 +0200160};
161
Jens Axboe932cc6d2007-06-21 13:10:21 +0200162/**
163 * splice_to_pipe - fill passed data into a pipe
164 * @pipe: pipe to fill
165 * @spd: data to fill
166 *
167 * Description:
Randy Dunlap79685b82007-07-27 08:08:51 +0200168 * @spd contains a map of pages and len/offset tuples, along with
Jens Axboe932cc6d2007-06-21 13:10:21 +0200169 * the struct pipe_buf_operations associated with these pages. This
170 * function will link that data to the pipe.
171 *
Jens Axboe83f91352006-04-02 23:05:09 +0200172 */
Jens Axboed6b29d72007-06-04 09:59:47 +0200173ssize_t splice_to_pipe(struct pipe_inode_info *pipe,
174 struct splice_pipe_desc *spd)
Jens Axboe5274f052006-03-30 15:15:30 +0200175{
Jens Axboe00de00b2007-06-15 13:14:22 +0200176 unsigned int spd_pages = spd->nr_pages;
Jens Axboe912d35f2006-04-26 10:59:21 +0200177 int ret, do_wakeup, page_nr;
Jens Axboe5274f052006-03-30 15:15:30 +0200178
179 ret = 0;
180 do_wakeup = 0;
Jens Axboe912d35f2006-04-26 10:59:21 +0200181 page_nr = 0;
Jens Axboe5274f052006-03-30 15:15:30 +0200182
Ingo Molnar3a326a22006-04-10 15:18:35 +0200183 if (pipe->inode)
184 mutex_lock(&pipe->inode->i_mutex);
Jens Axboe5274f052006-03-30 15:15:30 +0200185
Jens Axboe5274f052006-03-30 15:15:30 +0200186 for (;;) {
Ingo Molnar3a326a22006-04-10 15:18:35 +0200187 if (!pipe->readers) {
Jens Axboe5274f052006-03-30 15:15:30 +0200188 send_sig(SIGPIPE, current, 0);
189 if (!ret)
190 ret = -EPIPE;
191 break;
192 }
193
Jens Axboe6f767b02006-04-11 13:53:56 +0200194 if (pipe->nrbufs < PIPE_BUFFERS) {
195 int newbuf = (pipe->curbuf + pipe->nrbufs) & (PIPE_BUFFERS - 1);
Ingo Molnar3a326a22006-04-10 15:18:35 +0200196 struct pipe_buffer *buf = pipe->bufs + newbuf;
Jens Axboe5274f052006-03-30 15:15:30 +0200197
Jens Axboe912d35f2006-04-26 10:59:21 +0200198 buf->page = spd->pages[page_nr];
199 buf->offset = spd->partial[page_nr].offset;
200 buf->len = spd->partial[page_nr].len;
Jens Axboe497f9622007-06-11 12:00:45 +0200201 buf->private = spd->partial[page_nr].private;
Jens Axboe912d35f2006-04-26 10:59:21 +0200202 buf->ops = spd->ops;
Jens Axboe7afa6fd2006-05-01 20:02:33 +0200203 if (spd->flags & SPLICE_F_GIFT)
204 buf->flags |= PIPE_BUF_FLAG_GIFT;
205
Jens Axboe6f767b02006-04-11 13:53:56 +0200206 pipe->nrbufs++;
Jens Axboe912d35f2006-04-26 10:59:21 +0200207 page_nr++;
208 ret += buf->len;
209
Jens Axboe6f767b02006-04-11 13:53:56 +0200210 if (pipe->inode)
211 do_wakeup = 1;
Jens Axboe5274f052006-03-30 15:15:30 +0200212
Jens Axboe912d35f2006-04-26 10:59:21 +0200213 if (!--spd->nr_pages)
Jens Axboe5274f052006-03-30 15:15:30 +0200214 break;
Jens Axboe6f767b02006-04-11 13:53:56 +0200215 if (pipe->nrbufs < PIPE_BUFFERS)
Jens Axboe5274f052006-03-30 15:15:30 +0200216 continue;
217
218 break;
219 }
220
Jens Axboe912d35f2006-04-26 10:59:21 +0200221 if (spd->flags & SPLICE_F_NONBLOCK) {
Linus Torvalds29e35092006-04-02 12:46:35 -0700222 if (!ret)
223 ret = -EAGAIN;
224 break;
225 }
226
Jens Axboe5274f052006-03-30 15:15:30 +0200227 if (signal_pending(current)) {
228 if (!ret)
229 ret = -ERESTARTSYS;
230 break;
231 }
232
233 if (do_wakeup) {
Jens Axboec0bd1f62006-04-10 09:03:32 +0200234 smp_mb();
Ingo Molnar3a326a22006-04-10 15:18:35 +0200235 if (waitqueue_active(&pipe->wait))
236 wake_up_interruptible_sync(&pipe->wait);
237 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
Jens Axboe5274f052006-03-30 15:15:30 +0200238 do_wakeup = 0;
239 }
240
Ingo Molnar3a326a22006-04-10 15:18:35 +0200241 pipe->waiting_writers++;
242 pipe_wait(pipe);
243 pipe->waiting_writers--;
Jens Axboe5274f052006-03-30 15:15:30 +0200244 }
245
Jens Axboe02676e52007-06-15 13:16:13 +0200246 if (pipe->inode) {
Ingo Molnar3a326a22006-04-10 15:18:35 +0200247 mutex_unlock(&pipe->inode->i_mutex);
Jens Axboe5274f052006-03-30 15:15:30 +0200248
Jens Axboe02676e52007-06-15 13:16:13 +0200249 if (do_wakeup) {
250 smp_mb();
251 if (waitqueue_active(&pipe->wait))
252 wake_up_interruptible(&pipe->wait);
253 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
254 }
Jens Axboe5274f052006-03-30 15:15:30 +0200255 }
256
Jens Axboe00de00b2007-06-15 13:14:22 +0200257 while (page_nr < spd_pages)
Jens Axboebbdfc2f2007-11-06 23:29:47 -0800258 spd->spd_release(spd, page_nr++);
Jens Axboe5274f052006-03-30 15:15:30 +0200259
260 return ret;
261}
262
Jens Axboebbdfc2f2007-11-06 23:29:47 -0800263static void spd_release_page(struct splice_pipe_desc *spd, unsigned int i)
264{
265 page_cache_release(spd->pages[i]);
266}
267
Ingo Molnar3a326a22006-04-10 15:18:35 +0200268static int
Jens Axboecbb7e572006-04-11 14:57:50 +0200269__generic_file_splice_read(struct file *in, loff_t *ppos,
270 struct pipe_inode_info *pipe, size_t len,
271 unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +0200272{
273 struct address_space *mapping = in->f_mapping;
Fengguang Wud8983912007-07-19 01:48:06 -0700274 unsigned int loff, nr_pages, req_pages;
Jens Axboe16c523d2006-04-10 09:03:58 +0200275 struct page *pages[PIPE_BUFFERS];
Jens Axboe912d35f2006-04-26 10:59:21 +0200276 struct partial_page partial[PIPE_BUFFERS];
Jens Axboe5274f052006-03-30 15:15:30 +0200277 struct page *page;
Jens Axboe91ad66e2006-04-19 15:55:10 +0200278 pgoff_t index, end_index;
279 loff_t isize;
Jens Axboeeb207962006-04-27 11:05:22 +0200280 int error, page_nr;
Jens Axboe912d35f2006-04-26 10:59:21 +0200281 struct splice_pipe_desc spd = {
282 .pages = pages,
283 .partial = partial,
284 .flags = flags,
285 .ops = &page_cache_pipe_buf_ops,
Jens Axboebbdfc2f2007-11-06 23:29:47 -0800286 .spd_release = spd_release_page,
Jens Axboe912d35f2006-04-26 10:59:21 +0200287 };
Jens Axboe5274f052006-03-30 15:15:30 +0200288
Jens Axboecbb7e572006-04-11 14:57:50 +0200289 index = *ppos >> PAGE_CACHE_SHIFT;
Jens Axboe912d35f2006-04-26 10:59:21 +0200290 loff = *ppos & ~PAGE_CACHE_MASK;
Fengguang Wud8983912007-07-19 01:48:06 -0700291 req_pages = (len + loff + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
292 nr_pages = min(req_pages, (unsigned)PIPE_BUFFERS);
Jens Axboe5274f052006-03-30 15:15:30 +0200293
294 /*
Jens Axboeeb207962006-04-27 11:05:22 +0200295 * Lookup the (hopefully) full range of pages we need.
296 */
297 spd.nr_pages = find_get_pages_contig(mapping, index, nr_pages, pages);
Fengguang Wu431a48202007-07-19 01:48:05 -0700298 index += spd.nr_pages;
Jens Axboeeb207962006-04-27 11:05:22 +0200299
300 /*
301 * If find_get_pages_contig() returned fewer pages than we needed,
Fengguang Wu431a48202007-07-19 01:48:05 -0700302 * readahead/allocate the rest and fill in the holes.
Jens Axboeeb207962006-04-27 11:05:22 +0200303 */
Fengguang Wu431a48202007-07-19 01:48:05 -0700304 if (spd.nr_pages < nr_pages)
Rusty Russellcf914a72007-07-19 01:48:08 -0700305 page_cache_sync_readahead(mapping, &in->f_ra, in,
306 index, req_pages - spd.nr_pages);
Fengguang Wu431a48202007-07-19 01:48:05 -0700307
Jens Axboe932cc6d2007-06-21 13:10:21 +0200308 error = 0;
Jens Axboeeb207962006-04-27 11:05:22 +0200309 while (spd.nr_pages < nr_pages) {
310 /*
311 * Page could be there, find_get_pages_contig() breaks on
312 * the first hole.
313 */
314 page = find_get_page(mapping, index);
315 if (!page) {
Jens Axboee27dedd2006-05-01 19:59:54 +0200316 /*
Jens Axboeeb207962006-04-27 11:05:22 +0200317 * page didn't exist, allocate one.
318 */
319 page = page_cache_alloc_cold(mapping);
320 if (!page)
321 break;
322
323 error = add_to_page_cache_lru(page, mapping, index,
Hugh Dickins4cd13502008-04-03 23:35:22 +0100324 mapping_gfp_mask(mapping));
Jens Axboeeb207962006-04-27 11:05:22 +0200325 if (unlikely(error)) {
326 page_cache_release(page);
Jens Axboea0548872006-05-03 10:58:22 +0200327 if (error == -EEXIST)
328 continue;
Jens Axboeeb207962006-04-27 11:05:22 +0200329 break;
330 }
331 /*
332 * add_to_page_cache() locks the page, unlock it
333 * to avoid convoluting the logic below even more.
334 */
335 unlock_page(page);
336 }
337
338 pages[spd.nr_pages++] = page;
339 index++;
340 }
341
342 /*
343 * Now loop over the map and see if we need to start IO on any
344 * pages, fill in the partial map, etc.
345 */
346 index = *ppos >> PAGE_CACHE_SHIFT;
347 nr_pages = spd.nr_pages;
348 spd.nr_pages = 0;
349 for (page_nr = 0; page_nr < nr_pages; page_nr++) {
Jens Axboe82aa5d62006-04-20 13:05:48 +0200350 unsigned int this_len;
351
352 if (!len)
353 break;
354
355 /*
356 * this_len is the max we'll use from this page
357 */
Andrew Mortonba5f5d92006-04-25 15:33:34 +0200358 this_len = min_t(unsigned long, len, PAGE_CACHE_SIZE - loff);
Jens Axboeeb207962006-04-27 11:05:22 +0200359 page = pages[page_nr];
Jens Axboe7480a902006-04-11 13:52:47 +0200360
Fengguang Wua08a1662007-07-19 01:48:03 -0700361 if (PageReadahead(page))
Rusty Russellcf914a72007-07-19 01:48:08 -0700362 page_cache_async_readahead(mapping, &in->f_ra, in,
Fengguang Wud8983912007-07-19 01:48:06 -0700363 page, index, req_pages - page_nr);
Fengguang Wua08a1662007-07-19 01:48:03 -0700364
Jens Axboe7480a902006-04-11 13:52:47 +0200365 /*
366 * If the page isn't uptodate, we may need to start io on it
367 */
368 if (!PageUptodate(page)) {
Jens Axboec4f895c2006-04-19 15:56:12 +0200369 /*
370 * If in nonblock mode then dont block on waiting
371 * for an in-flight io page
372 */
Fengguang Wu9ae9d682007-05-08 08:44:36 +0200373 if (flags & SPLICE_F_NONBLOCK) {
Nick Piggin529ae9a2008-08-02 12:01:03 +0200374 if (!trylock_page(page)) {
Jens Axboe8191ecd2008-04-10 08:24:25 +0200375 error = -EAGAIN;
Fengguang Wu9ae9d682007-05-08 08:44:36 +0200376 break;
Jens Axboe8191ecd2008-04-10 08:24:25 +0200377 }
Fengguang Wu9ae9d682007-05-08 08:44:36 +0200378 } else
379 lock_page(page);
Jens Axboe7480a902006-04-11 13:52:47 +0200380
381 /*
Miklos Szeredi32502b82008-07-04 09:35:17 +0200382 * Page was truncated, or invalidated by the
383 * filesystem. Redo the find/create, but this time the
384 * page is kept locked, so there's no chance of another
385 * race with truncate/invalidate.
Jens Axboe7480a902006-04-11 13:52:47 +0200386 */
387 if (!page->mapping) {
388 unlock_page(page);
Miklos Szeredi32502b82008-07-04 09:35:17 +0200389 page = find_or_create_page(mapping, index,
390 mapping_gfp_mask(mapping));
391
392 if (!page) {
393 error = -ENOMEM;
394 break;
395 }
396 page_cache_release(pages[page_nr]);
397 pages[page_nr] = page;
Jens Axboe7480a902006-04-11 13:52:47 +0200398 }
399 /*
400 * page was already under io and is now done, great
401 */
402 if (PageUptodate(page)) {
403 unlock_page(page);
404 goto fill_it;
405 }
406
Jens Axboe7480a902006-04-11 13:52:47 +0200407 /*
408 * need to read in the page
409 */
410 error = mapping->a_ops->readpage(in, page);
Jens Axboe7480a902006-04-11 13:52:47 +0200411 if (unlikely(error)) {
Jens Axboeeb207962006-04-27 11:05:22 +0200412 /*
413 * We really should re-lookup the page here,
414 * but it complicates things a lot. Instead
415 * lets just do what we already stored, and
416 * we'll get it the next time we are called.
417 */
Jens Axboe7480a902006-04-11 13:52:47 +0200418 if (error == AOP_TRUNCATED_PAGE)
Jens Axboeeb207962006-04-27 11:05:22 +0200419 error = 0;
420
Jens Axboe7480a902006-04-11 13:52:47 +0200421 break;
422 }
Jens Axboe620a3242007-06-07 09:39:42 +0200423 }
424fill_it:
425 /*
426 * i_size must be checked after PageUptodate.
427 */
428 isize = i_size_read(mapping->host);
429 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
430 if (unlikely(!isize || index > end_index))
431 break;
432
433 /*
434 * if this is the last page, see if we need to shrink
435 * the length and stop
436 */
437 if (end_index == index) {
438 unsigned int plen;
Jens Axboe91ad66e2006-04-19 15:55:10 +0200439
440 /*
Jens Axboe620a3242007-06-07 09:39:42 +0200441 * max good bytes in this page
Jens Axboe91ad66e2006-04-19 15:55:10 +0200442 */
Jens Axboe620a3242007-06-07 09:39:42 +0200443 plen = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
444 if (plen <= loff)
Jens Axboe91ad66e2006-04-19 15:55:10 +0200445 break;
Jens Axboe91ad66e2006-04-19 15:55:10 +0200446
447 /*
Jens Axboe620a3242007-06-07 09:39:42 +0200448 * force quit after adding this page
Jens Axboe91ad66e2006-04-19 15:55:10 +0200449 */
Jens Axboe620a3242007-06-07 09:39:42 +0200450 this_len = min(this_len, plen - loff);
451 len = this_len;
Jens Axboe7480a902006-04-11 13:52:47 +0200452 }
Jens Axboe620a3242007-06-07 09:39:42 +0200453
Jens Axboeeb207962006-04-27 11:05:22 +0200454 partial[page_nr].offset = loff;
455 partial[page_nr].len = this_len;
Jens Axboe82aa5d62006-04-20 13:05:48 +0200456 len -= this_len;
Jens Axboe91ad66e2006-04-19 15:55:10 +0200457 loff = 0;
Jens Axboeeb207962006-04-27 11:05:22 +0200458 spd.nr_pages++;
459 index++;
Jens Axboe5274f052006-03-30 15:15:30 +0200460 }
461
Jens Axboeeb207962006-04-27 11:05:22 +0200462 /*
Hugh Dickins475ecad2007-06-07 09:36:00 +0200463 * Release any pages at the end, if we quit early. 'page_nr' is how far
Jens Axboeeb207962006-04-27 11:05:22 +0200464 * we got, 'nr_pages' is how many pages are in the map.
465 */
466 while (page_nr < nr_pages)
467 page_cache_release(pages[page_nr++]);
Fengguang Wuf4e6b492007-10-16 01:24:33 -0700468 in->f_ra.prev_pos = (loff_t)index << PAGE_CACHE_SHIFT;
Jens Axboeeb207962006-04-27 11:05:22 +0200469
Jens Axboe912d35f2006-04-26 10:59:21 +0200470 if (spd.nr_pages)
Jens Axboe00522fb2006-04-26 14:39:29 +0200471 return splice_to_pipe(pipe, &spd);
Jens Axboe5274f052006-03-30 15:15:30 +0200472
Jens Axboe7480a902006-04-11 13:52:47 +0200473 return error;
Jens Axboe5274f052006-03-30 15:15:30 +0200474}
475
Jens Axboe83f91352006-04-02 23:05:09 +0200476/**
477 * generic_file_splice_read - splice data from file to a pipe
478 * @in: file to splice from
Jens Axboe932cc6d2007-06-21 13:10:21 +0200479 * @ppos: position in @in
Jens Axboe83f91352006-04-02 23:05:09 +0200480 * @pipe: pipe to splice to
481 * @len: number of bytes to splice
482 * @flags: splice modifier flags
483 *
Jens Axboe932cc6d2007-06-21 13:10:21 +0200484 * Description:
485 * Will read pages from given file and fill them into a pipe. Can be
486 * used as long as the address_space operations for the source implements
487 * a readpage() hook.
488 *
Jens Axboe83f91352006-04-02 23:05:09 +0200489 */
Jens Axboecbb7e572006-04-11 14:57:50 +0200490ssize_t generic_file_splice_read(struct file *in, loff_t *ppos,
491 struct pipe_inode_info *pipe, size_t len,
492 unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +0200493{
Jens Axboed366d3982007-06-01 14:54:11 +0200494 loff_t isize, left;
Jens Axboe8191ecd2008-04-10 08:24:25 +0200495 int ret;
Jens Axboed366d3982007-06-01 14:54:11 +0200496
497 isize = i_size_read(in->f_mapping->host);
498 if (unlikely(*ppos >= isize))
499 return 0;
500
501 left = isize - *ppos;
502 if (unlikely(left < len))
503 len = left;
Jens Axboe5274f052006-03-30 15:15:30 +0200504
Jens Axboe8191ecd2008-04-10 08:24:25 +0200505 ret = __generic_file_splice_read(in, ppos, pipe, len, flags);
506 if (ret > 0)
Jens Axboecbb7e572006-04-11 14:57:50 +0200507 *ppos += ret;
Jens Axboe5274f052006-03-30 15:15:30 +0200508
509 return ret;
510}
511
Jens Axboe059a8f32006-04-02 23:06:05 +0200512EXPORT_SYMBOL(generic_file_splice_read);
513
Jens Axboe5274f052006-03-30 15:15:30 +0200514/*
Jens Axboe4f6f0bd2006-04-02 23:04:46 +0200515 * Send 'sd->len' bytes to socket from 'sd->file' at position 'sd->pos'
Jens Axboe016b6612006-04-25 15:42:00 +0200516 * using sendpage(). Return the number of bytes sent.
Jens Axboe5274f052006-03-30 15:15:30 +0200517 */
Jens Axboe76ad4d12006-05-03 10:41:33 +0200518static int pipe_to_sendpage(struct pipe_inode_info *pipe,
Jens Axboe5274f052006-03-30 15:15:30 +0200519 struct pipe_buffer *buf, struct splice_desc *sd)
520{
Jens Axboe6a14b902007-06-14 13:08:55 +0200521 struct file *file = sd->u.file;
Jens Axboe5274f052006-03-30 15:15:30 +0200522 loff_t pos = sd->pos;
Jens Axboef84d7512006-05-01 19:59:03 +0200523 int ret, more;
Jens Axboe5274f052006-03-30 15:15:30 +0200524
Jens Axboecac36bb02007-06-14 13:10:48 +0200525 ret = buf->ops->confirm(pipe, buf);
Jens Axboef84d7512006-05-01 19:59:03 +0200526 if (!ret) {
527 more = (sd->flags & SPLICE_F_MORE) || sd->len < sd->total_len;
Jens Axboe5274f052006-03-30 15:15:30 +0200528
Jens Axboef84d7512006-05-01 19:59:03 +0200529 ret = file->f_op->sendpage(file, buf->page, buf->offset,
530 sd->len, &pos, more);
531 }
Jens Axboe5274f052006-03-30 15:15:30 +0200532
Jens Axboe016b6612006-04-25 15:42:00 +0200533 return ret;
Jens Axboe5274f052006-03-30 15:15:30 +0200534}
535
536/*
537 * This is a little more tricky than the file -> pipe splicing. There are
538 * basically three cases:
539 *
540 * - Destination page already exists in the address space and there
541 * are users of it. For that case we have no other option that
542 * copying the data. Tough luck.
543 * - Destination page already exists in the address space, but there
544 * are no users of it. Make sure it's uptodate, then drop it. Fall
545 * through to last case.
546 * - Destination page does not exist, we can add the pipe page to
547 * the page cache and avoid the copy.
548 *
Jens Axboe83f91352006-04-02 23:05:09 +0200549 * If asked to move pages to the output file (SPLICE_F_MOVE is set in
550 * sd->flags), we attempt to migrate pages from the pipe to the output
551 * file address space page cache. This is possible if no one else has
552 * the pipe page referenced outside of the pipe and page cache. If
553 * SPLICE_F_MOVE isn't set, or we cannot move the page, we simply create
554 * a new page in the output file page cache and fill/dirty that.
Jens Axboe5274f052006-03-30 15:15:30 +0200555 */
Jens Axboe76ad4d12006-05-03 10:41:33 +0200556static int pipe_to_file(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
Jens Axboe5274f052006-03-30 15:15:30 +0200557 struct splice_desc *sd)
558{
Jens Axboe6a14b902007-06-14 13:08:55 +0200559 struct file *file = sd->u.file;
Jens Axboe5274f052006-03-30 15:15:30 +0200560 struct address_space *mapping = file->f_mapping;
Jens Axboe016b6612006-04-25 15:42:00 +0200561 unsigned int offset, this_len;
Jens Axboe5274f052006-03-30 15:15:30 +0200562 struct page *page;
Nick Pigginafddba42007-10-16 01:25:01 -0700563 void *fsdata;
Jens Axboe3e7ee3e2006-04-02 23:11:04 +0200564 int ret;
Jens Axboe5274f052006-03-30 15:15:30 +0200565
566 /*
Jens Axboe49d0b212006-04-10 09:04:41 +0200567 * make sure the data in this buffer is uptodate
Jens Axboe5274f052006-03-30 15:15:30 +0200568 */
Jens Axboecac36bb02007-06-14 13:10:48 +0200569 ret = buf->ops->confirm(pipe, buf);
Jens Axboef84d7512006-05-01 19:59:03 +0200570 if (unlikely(ret))
571 return ret;
Jens Axboe5274f052006-03-30 15:15:30 +0200572
Jens Axboe5274f052006-03-30 15:15:30 +0200573 offset = sd->pos & ~PAGE_CACHE_MASK;
574
Jens Axboe016b6612006-04-25 15:42:00 +0200575 this_len = sd->len;
576 if (this_len + offset > PAGE_CACHE_SIZE)
577 this_len = PAGE_CACHE_SIZE - offset;
578
Nick Pigginafddba42007-10-16 01:25:01 -0700579 ret = pagecache_write_begin(file, mapping, sd->pos, this_len,
580 AOP_FLAG_UNINTERRUPTIBLE, &page, &fsdata);
581 if (unlikely(ret))
582 goto out;
Jens Axboe5274f052006-03-30 15:15:30 +0200583
Jens Axboe0568b402006-05-01 19:50:48 +0200584 if (buf->page != page) {
Jens Axboef84d7512006-05-01 19:59:03 +0200585 /*
586 * Careful, ->map() uses KM_USER0!
587 */
Jens Axboe76ad4d12006-05-03 10:41:33 +0200588 char *src = buf->ops->map(pipe, buf, 1);
Jens Axboef84d7512006-05-01 19:59:03 +0200589 char *dst = kmap_atomic(page, KM_USER1);
Jens Axboe5abc97a2006-03-30 15:16:46 +0200590
Jens Axboe016b6612006-04-25 15:42:00 +0200591 memcpy(dst + offset, src + buf->offset, this_len);
Jens Axboe5abc97a2006-03-30 15:16:46 +0200592 flush_dcache_page(page);
Jens Axboef84d7512006-05-01 19:59:03 +0200593 kunmap_atomic(dst, KM_USER1);
Jens Axboe76ad4d12006-05-03 10:41:33 +0200594 buf->ops->unmap(pipe, buf, src);
Jens Axboe5abc97a2006-03-30 15:16:46 +0200595 }
Nick Pigginafddba42007-10-16 01:25:01 -0700596 ret = pagecache_write_end(file, mapping, sd->pos, this_len, this_len,
597 page, fsdata);
Jens Axboe5274f052006-03-30 15:15:30 +0200598out:
Jens Axboe5274f052006-03-30 15:15:30 +0200599 return ret;
600}
601
Jens Axboe932cc6d2007-06-21 13:10:21 +0200602/**
603 * __splice_from_pipe - splice data from a pipe to given actor
604 * @pipe: pipe to splice from
605 * @sd: information to @actor
606 * @actor: handler that splices the data
607 *
608 * Description:
609 * This function does little more than loop over the pipe and call
610 * @actor to do the actual moving of a single struct pipe_buffer to
611 * the desired destination. See pipe_to_file, pipe_to_sendpage, or
612 * pipe_to_user.
613 *
Jens Axboe83f91352006-04-02 23:05:09 +0200614 */
Jens Axboec66ab6f2007-06-12 21:17:17 +0200615ssize_t __splice_from_pipe(struct pipe_inode_info *pipe, struct splice_desc *sd,
616 splice_actor *actor)
Jens Axboe5274f052006-03-30 15:15:30 +0200617{
Jens Axboe5274f052006-03-30 15:15:30 +0200618 int ret, do_wakeup, err;
Jens Axboe5274f052006-03-30 15:15:30 +0200619
620 ret = 0;
621 do_wakeup = 0;
622
Jens Axboe5274f052006-03-30 15:15:30 +0200623 for (;;) {
Jens Axboe6f767b02006-04-11 13:53:56 +0200624 if (pipe->nrbufs) {
625 struct pipe_buffer *buf = pipe->bufs + pipe->curbuf;
Eric Dumazetd4c3cca2006-12-13 00:34:04 -0800626 const struct pipe_buf_operations *ops = buf->ops;
Jens Axboe5274f052006-03-30 15:15:30 +0200627
Jens Axboec66ab6f2007-06-12 21:17:17 +0200628 sd->len = buf->len;
629 if (sd->len > sd->total_len)
630 sd->len = sd->total_len;
Jens Axboe5274f052006-03-30 15:15:30 +0200631
Jens Axboec66ab6f2007-06-12 21:17:17 +0200632 err = actor(pipe, buf, sd);
Jens Axboe016b6612006-04-25 15:42:00 +0200633 if (err <= 0) {
Jens Axboe5274f052006-03-30 15:15:30 +0200634 if (!ret && err != -ENODATA)
635 ret = err;
636
637 break;
638 }
639
Jens Axboe016b6612006-04-25 15:42:00 +0200640 ret += err;
641 buf->offset += err;
642 buf->len -= err;
643
Jens Axboec66ab6f2007-06-12 21:17:17 +0200644 sd->len -= err;
645 sd->pos += err;
646 sd->total_len -= err;
647 if (sd->len)
Jens Axboe016b6612006-04-25 15:42:00 +0200648 continue;
Ingo Molnar73d62d82006-04-11 13:57:21 +0200649
Jens Axboe5274f052006-03-30 15:15:30 +0200650 if (!buf->len) {
651 buf->ops = NULL;
Ingo Molnar3a326a22006-04-10 15:18:35 +0200652 ops->release(pipe, buf);
Jens Axboe6f767b02006-04-11 13:53:56 +0200653 pipe->curbuf = (pipe->curbuf + 1) & (PIPE_BUFFERS - 1);
654 pipe->nrbufs--;
655 if (pipe->inode)
656 do_wakeup = 1;
Jens Axboe5274f052006-03-30 15:15:30 +0200657 }
658
Jens Axboec66ab6f2007-06-12 21:17:17 +0200659 if (!sd->total_len)
Jens Axboe5274f052006-03-30 15:15:30 +0200660 break;
661 }
662
Jens Axboe6f767b02006-04-11 13:53:56 +0200663 if (pipe->nrbufs)
Jens Axboe5274f052006-03-30 15:15:30 +0200664 continue;
Ingo Molnar3a326a22006-04-10 15:18:35 +0200665 if (!pipe->writers)
Jens Axboe5274f052006-03-30 15:15:30 +0200666 break;
Ingo Molnar3a326a22006-04-10 15:18:35 +0200667 if (!pipe->waiting_writers) {
Jens Axboe5274f052006-03-30 15:15:30 +0200668 if (ret)
669 break;
670 }
671
Jens Axboec66ab6f2007-06-12 21:17:17 +0200672 if (sd->flags & SPLICE_F_NONBLOCK) {
Linus Torvalds29e35092006-04-02 12:46:35 -0700673 if (!ret)
674 ret = -EAGAIN;
675 break;
676 }
677
Jens Axboe5274f052006-03-30 15:15:30 +0200678 if (signal_pending(current)) {
679 if (!ret)
680 ret = -ERESTARTSYS;
681 break;
682 }
683
684 if (do_wakeup) {
Jens Axboec0bd1f62006-04-10 09:03:32 +0200685 smp_mb();
Ingo Molnar3a326a22006-04-10 15:18:35 +0200686 if (waitqueue_active(&pipe->wait))
687 wake_up_interruptible_sync(&pipe->wait);
688 kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
Jens Axboe5274f052006-03-30 15:15:30 +0200689 do_wakeup = 0;
690 }
691
Ingo Molnar3a326a22006-04-10 15:18:35 +0200692 pipe_wait(pipe);
Jens Axboe5274f052006-03-30 15:15:30 +0200693 }
694
Jens Axboe5274f052006-03-30 15:15:30 +0200695 if (do_wakeup) {
Jens Axboec0bd1f62006-04-10 09:03:32 +0200696 smp_mb();
Ingo Molnar3a326a22006-04-10 15:18:35 +0200697 if (waitqueue_active(&pipe->wait))
698 wake_up_interruptible(&pipe->wait);
699 kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
Jens Axboe5274f052006-03-30 15:15:30 +0200700 }
701
Jens Axboe5274f052006-03-30 15:15:30 +0200702 return ret;
Jens Axboe5274f052006-03-30 15:15:30 +0200703}
Mark Fasheh40bee44e2007-03-21 13:11:02 +0100704EXPORT_SYMBOL(__splice_from_pipe);
Jens Axboe5274f052006-03-30 15:15:30 +0200705
Jens Axboe932cc6d2007-06-21 13:10:21 +0200706/**
707 * splice_from_pipe - splice data from a pipe to a file
708 * @pipe: pipe to splice from
709 * @out: file to splice to
710 * @ppos: position in @out
711 * @len: how many bytes to splice
712 * @flags: splice modifier flags
713 * @actor: handler that splices the data
714 *
715 * Description:
716 * See __splice_from_pipe. This function locks the input and output inodes,
717 * otherwise it's identical to __splice_from_pipe().
718 *
719 */
Mark Fasheh6da61802006-10-17 18:43:07 +0200720ssize_t splice_from_pipe(struct pipe_inode_info *pipe, struct file *out,
721 loff_t *ppos, size_t len, unsigned int flags,
722 splice_actor *actor)
723{
724 ssize_t ret;
725 struct inode *inode = out->f_mapping->host;
Jens Axboec66ab6f2007-06-12 21:17:17 +0200726 struct splice_desc sd = {
727 .total_len = len,
728 .flags = flags,
729 .pos = *ppos,
Jens Axboe6a14b902007-06-14 13:08:55 +0200730 .u.file = out,
Jens Axboec66ab6f2007-06-12 21:17:17 +0200731 };
Mark Fasheh6da61802006-10-17 18:43:07 +0200732
733 /*
734 * The actor worker might be calling ->prepare_write and
735 * ->commit_write. Most of the time, these expect i_mutex to
736 * be held. Since this may result in an ABBA deadlock with
737 * pipe->inode, we have to order lock acquiry here.
738 */
739 inode_double_lock(inode, pipe->inode);
Jens Axboec66ab6f2007-06-12 21:17:17 +0200740 ret = __splice_from_pipe(pipe, &sd, actor);
Mark Fasheh6da61802006-10-17 18:43:07 +0200741 inode_double_unlock(inode, pipe->inode);
742
743 return ret;
744}
745
746/**
747 * generic_file_splice_write_nolock - generic_file_splice_write without mutexes
748 * @pipe: pipe info
749 * @out: file to write to
Jens Axboe932cc6d2007-06-21 13:10:21 +0200750 * @ppos: position in @out
Mark Fasheh6da61802006-10-17 18:43:07 +0200751 * @len: number of bytes to splice
752 * @flags: splice modifier flags
753 *
Jens Axboe932cc6d2007-06-21 13:10:21 +0200754 * Description:
755 * Will either move or copy pages (determined by @flags options) from
756 * the given pipe inode to the given file. The caller is responsible
757 * for acquiring i_mutex on both inodes.
Mark Fasheh6da61802006-10-17 18:43:07 +0200758 *
759 */
760ssize_t
761generic_file_splice_write_nolock(struct pipe_inode_info *pipe, struct file *out,
762 loff_t *ppos, size_t len, unsigned int flags)
763{
764 struct address_space *mapping = out->f_mapping;
765 struct inode *inode = mapping->host;
Jens Axboec66ab6f2007-06-12 21:17:17 +0200766 struct splice_desc sd = {
767 .total_len = len,
768 .flags = flags,
769 .pos = *ppos,
Jens Axboe6a14b902007-06-14 13:08:55 +0200770 .u.file = out,
Jens Axboec66ab6f2007-06-12 21:17:17 +0200771 };
Mark Fasheh6da61802006-10-17 18:43:07 +0200772 ssize_t ret;
773 int err;
774
Miklos Szeredi2f1936b2008-06-24 16:50:14 +0200775 err = file_remove_suid(out);
Jens Axboe8c34e2d2006-10-17 19:43:22 +0200776 if (unlikely(err))
777 return err;
778
Jens Axboec66ab6f2007-06-12 21:17:17 +0200779 ret = __splice_from_pipe(pipe, &sd, pipe_to_file);
Mark Fasheh6da61802006-10-17 18:43:07 +0200780 if (ret > 0) {
Jens Axboe17ee4f42007-06-15 13:10:37 +0200781 unsigned long nr_pages;
782
Mark Fasheh6da61802006-10-17 18:43:07 +0200783 *ppos += ret;
Jens Axboe17ee4f42007-06-15 13:10:37 +0200784 nr_pages = (ret + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
Mark Fasheh6da61802006-10-17 18:43:07 +0200785
786 /*
787 * If file or inode is SYNC and we actually wrote some data,
788 * sync it.
789 */
790 if (unlikely((out->f_flags & O_SYNC) || IS_SYNC(inode))) {
791 err = generic_osync_inode(inode, mapping,
792 OSYNC_METADATA|OSYNC_DATA);
793
794 if (err)
795 ret = err;
796 }
Jens Axboe17ee4f42007-06-15 13:10:37 +0200797 balance_dirty_pages_ratelimited_nr(mapping, nr_pages);
Mark Fasheh6da61802006-10-17 18:43:07 +0200798 }
799
800 return ret;
801}
802
803EXPORT_SYMBOL(generic_file_splice_write_nolock);
804
Jens Axboe83f91352006-04-02 23:05:09 +0200805/**
806 * generic_file_splice_write - splice data from a pipe to a file
Ingo Molnar3a326a22006-04-10 15:18:35 +0200807 * @pipe: pipe info
Jens Axboe83f91352006-04-02 23:05:09 +0200808 * @out: file to write to
Jens Axboe932cc6d2007-06-21 13:10:21 +0200809 * @ppos: position in @out
Jens Axboe83f91352006-04-02 23:05:09 +0200810 * @len: number of bytes to splice
811 * @flags: splice modifier flags
812 *
Jens Axboe932cc6d2007-06-21 13:10:21 +0200813 * Description:
814 * Will either move or copy pages (determined by @flags options) from
815 * the given pipe inode to the given file.
Jens Axboe83f91352006-04-02 23:05:09 +0200816 *
817 */
Ingo Molnar3a326a22006-04-10 15:18:35 +0200818ssize_t
819generic_file_splice_write(struct pipe_inode_info *pipe, struct file *out,
Jens Axboecbb7e572006-04-11 14:57:50 +0200820 loff_t *ppos, size_t len, unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +0200821{
Jens Axboe4f6f0bd2006-04-02 23:04:46 +0200822 struct address_space *mapping = out->f_mapping;
Jens Axboe8c34e2d2006-10-17 19:43:22 +0200823 struct inode *inode = mapping->host;
Miklos Szeredi7f3d4ee2008-05-07 09:22:39 +0200824 struct splice_desc sd = {
825 .total_len = len,
826 .flags = flags,
827 .pos = *ppos,
828 .u.file = out,
829 };
Ingo Molnar3a326a22006-04-10 15:18:35 +0200830 ssize_t ret;
Jens Axboe8c34e2d2006-10-17 19:43:22 +0200831
Miklos Szeredi7f3d4ee2008-05-07 09:22:39 +0200832 inode_double_lock(inode, pipe->inode);
Miklos Szeredi2f1936b2008-06-24 16:50:14 +0200833 ret = file_remove_suid(out);
Miklos Szeredi7f3d4ee2008-05-07 09:22:39 +0200834 if (likely(!ret))
835 ret = __splice_from_pipe(pipe, &sd, pipe_to_file);
836 inode_double_unlock(inode, pipe->inode);
Jens Axboea4514eb2006-04-19 15:57:05 +0200837 if (ret > 0) {
Jens Axboe17ee4f42007-06-15 13:10:37 +0200838 unsigned long nr_pages;
839
Jens Axboea4514eb2006-04-19 15:57:05 +0200840 *ppos += ret;
Jens Axboe17ee4f42007-06-15 13:10:37 +0200841 nr_pages = (ret + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
Jens Axboe4f6f0bd2006-04-02 23:04:46 +0200842
Jens Axboea4514eb2006-04-19 15:57:05 +0200843 /*
844 * If file or inode is SYNC and we actually wrote some data,
845 * sync it.
846 */
847 if (unlikely((out->f_flags & O_SYNC) || IS_SYNC(inode))) {
Miklos Szeredi7f3d4ee2008-05-07 09:22:39 +0200848 int err;
849
Jens Axboea4514eb2006-04-19 15:57:05 +0200850 mutex_lock(&inode->i_mutex);
851 err = generic_osync_inode(inode, mapping,
852 OSYNC_METADATA|OSYNC_DATA);
853 mutex_unlock(&inode->i_mutex);
854
855 if (err)
856 ret = err;
857 }
Jens Axboe17ee4f42007-06-15 13:10:37 +0200858 balance_dirty_pages_ratelimited_nr(mapping, nr_pages);
Jens Axboe4f6f0bd2006-04-02 23:04:46 +0200859 }
860
861 return ret;
Jens Axboe5274f052006-03-30 15:15:30 +0200862}
863
Jens Axboe059a8f32006-04-02 23:06:05 +0200864EXPORT_SYMBOL(generic_file_splice_write);
865
Jens Axboe83f91352006-04-02 23:05:09 +0200866/**
867 * generic_splice_sendpage - splice data from a pipe to a socket
Jens Axboe932cc6d2007-06-21 13:10:21 +0200868 * @pipe: pipe to splice from
Jens Axboe83f91352006-04-02 23:05:09 +0200869 * @out: socket to write to
Jens Axboe932cc6d2007-06-21 13:10:21 +0200870 * @ppos: position in @out
Jens Axboe83f91352006-04-02 23:05:09 +0200871 * @len: number of bytes to splice
872 * @flags: splice modifier flags
873 *
Jens Axboe932cc6d2007-06-21 13:10:21 +0200874 * Description:
875 * Will send @len bytes from the pipe to a network socket. No data copying
876 * is involved.
Jens Axboe83f91352006-04-02 23:05:09 +0200877 *
878 */
Ingo Molnar3a326a22006-04-10 15:18:35 +0200879ssize_t generic_splice_sendpage(struct pipe_inode_info *pipe, struct file *out,
Jens Axboecbb7e572006-04-11 14:57:50 +0200880 loff_t *ppos, size_t len, unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +0200881{
Jens Axboe00522fb2006-04-26 14:39:29 +0200882 return splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_sendpage);
Jens Axboe5274f052006-03-30 15:15:30 +0200883}
884
Jens Axboe059a8f32006-04-02 23:06:05 +0200885EXPORT_SYMBOL(generic_splice_sendpage);
Jeff Garzika0f06782006-03-30 23:06:13 -0500886
Jens Axboe83f91352006-04-02 23:05:09 +0200887/*
888 * Attempt to initiate a splice from pipe to file.
889 */
Ingo Molnar3a326a22006-04-10 15:18:35 +0200890static long do_splice_from(struct pipe_inode_info *pipe, struct file *out,
Jens Axboecbb7e572006-04-11 14:57:50 +0200891 loff_t *ppos, size_t len, unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +0200892{
Jens Axboe5274f052006-03-30 15:15:30 +0200893 int ret;
894
Jens Axboe49570e92006-04-11 13:56:09 +0200895 if (unlikely(!out->f_op || !out->f_op->splice_write))
Jens Axboe5274f052006-03-30 15:15:30 +0200896 return -EINVAL;
897
Jens Axboe49570e92006-04-11 13:56:09 +0200898 if (unlikely(!(out->f_mode & FMODE_WRITE)))
Jens Axboe5274f052006-03-30 15:15:30 +0200899 return -EBADF;
900
Linus Torvaldsefc968d2008-10-09 14:04:54 -0700901 if (unlikely(out->f_flags & O_APPEND))
902 return -EINVAL;
903
Jens Axboecbb7e572006-04-11 14:57:50 +0200904 ret = rw_verify_area(WRITE, out, ppos, len);
Jens Axboe5274f052006-03-30 15:15:30 +0200905 if (unlikely(ret < 0))
906 return ret;
907
Jens Axboecbb7e572006-04-11 14:57:50 +0200908 return out->f_op->splice_write(pipe, out, ppos, len, flags);
Jens Axboe5274f052006-03-30 15:15:30 +0200909}
910
Jens Axboe83f91352006-04-02 23:05:09 +0200911/*
912 * Attempt to initiate a splice from a file to a pipe.
913 */
Jens Axboecbb7e572006-04-11 14:57:50 +0200914static long do_splice_to(struct file *in, loff_t *ppos,
915 struct pipe_inode_info *pipe, size_t len,
916 unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +0200917{
Jens Axboe5274f052006-03-30 15:15:30 +0200918 int ret;
919
Jens Axboe49570e92006-04-11 13:56:09 +0200920 if (unlikely(!in->f_op || !in->f_op->splice_read))
Jens Axboe5274f052006-03-30 15:15:30 +0200921 return -EINVAL;
922
Jens Axboe49570e92006-04-11 13:56:09 +0200923 if (unlikely(!(in->f_mode & FMODE_READ)))
Jens Axboe5274f052006-03-30 15:15:30 +0200924 return -EBADF;
925
Jens Axboecbb7e572006-04-11 14:57:50 +0200926 ret = rw_verify_area(READ, in, ppos, len);
Jens Axboe5274f052006-03-30 15:15:30 +0200927 if (unlikely(ret < 0))
928 return ret;
929
Jens Axboecbb7e572006-04-11 14:57:50 +0200930 return in->f_op->splice_read(in, ppos, pipe, len, flags);
Jens Axboe5274f052006-03-30 15:15:30 +0200931}
932
Jens Axboe932cc6d2007-06-21 13:10:21 +0200933/**
934 * splice_direct_to_actor - splices data directly between two non-pipes
935 * @in: file to splice from
936 * @sd: actor information on where to splice to
937 * @actor: handles the data splicing
938 *
939 * Description:
940 * This is a special case helper to splice directly between two
941 * points, without requiring an explicit pipe. Internally an allocated
Randy Dunlap79685b82007-07-27 08:08:51 +0200942 * pipe is cached in the process, and reused during the lifetime of
Jens Axboe932cc6d2007-06-21 13:10:21 +0200943 * that process.
944 *
Jens Axboec66ab6f2007-06-12 21:17:17 +0200945 */
946ssize_t splice_direct_to_actor(struct file *in, struct splice_desc *sd,
947 splice_direct_actor *actor)
Jens Axboeb92ce552006-04-11 13:52:07 +0200948{
949 struct pipe_inode_info *pipe;
950 long ret, bytes;
951 umode_t i_mode;
Jens Axboec66ab6f2007-06-12 21:17:17 +0200952 size_t len;
953 int i, flags;
Jens Axboeb92ce552006-04-11 13:52:07 +0200954
955 /*
956 * We require the input being a regular file, as we don't want to
957 * randomly drop data for eg socket -> socket splicing. Use the
958 * piped splicing for that!
959 */
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800960 i_mode = in->f_path.dentry->d_inode->i_mode;
Jens Axboeb92ce552006-04-11 13:52:07 +0200961 if (unlikely(!S_ISREG(i_mode) && !S_ISBLK(i_mode)))
962 return -EINVAL;
963
964 /*
965 * neither in nor out is a pipe, setup an internal pipe attached to
966 * 'out' and transfer the wanted data from 'in' to 'out' through that
967 */
968 pipe = current->splice_pipe;
Jens Axboe49570e92006-04-11 13:56:09 +0200969 if (unlikely(!pipe)) {
Jens Axboeb92ce552006-04-11 13:52:07 +0200970 pipe = alloc_pipe_info(NULL);
971 if (!pipe)
972 return -ENOMEM;
973
974 /*
975 * We don't have an immediate reader, but we'll read the stuff
Jens Axboe00522fb2006-04-26 14:39:29 +0200976 * out of the pipe right after the splice_to_pipe(). So set
Jens Axboeb92ce552006-04-11 13:52:07 +0200977 * PIPE_READERS appropriately.
978 */
979 pipe->readers = 1;
980
981 current->splice_pipe = pipe;
982 }
983
984 /*
Ingo Molnar73d62d82006-04-11 13:57:21 +0200985 * Do the splice.
Jens Axboeb92ce552006-04-11 13:52:07 +0200986 */
987 ret = 0;
988 bytes = 0;
Jens Axboec66ab6f2007-06-12 21:17:17 +0200989 len = sd->total_len;
990 flags = sd->flags;
991
992 /*
993 * Don't block on output, we have to drain the direct pipe.
994 */
995 sd->flags &= ~SPLICE_F_NONBLOCK;
Jens Axboeb92ce552006-04-11 13:52:07 +0200996
997 while (len) {
Jens Axboe51a92c02007-07-13 14:11:43 +0200998 size_t read_len;
Tom Zanussia82c53a2008-05-09 13:28:36 +0200999 loff_t pos = sd->pos, prev_pos = pos;
Jens Axboeb92ce552006-04-11 13:52:07 +02001000
Jens Axboebcd4f3a2007-07-16 14:41:49 +02001001 ret = do_splice_to(in, &pos, pipe, len, flags);
Jens Axboe51a92c02007-07-13 14:11:43 +02001002 if (unlikely(ret <= 0))
Jens Axboeb92ce552006-04-11 13:52:07 +02001003 goto out_release;
1004
1005 read_len = ret;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001006 sd->total_len = read_len;
Jens Axboeb92ce552006-04-11 13:52:07 +02001007
1008 /*
1009 * NOTE: nonblocking mode only applies to the input. We
1010 * must not do the output in nonblocking mode as then we
1011 * could get stuck data in the internal pipe:
1012 */
Jens Axboec66ab6f2007-06-12 21:17:17 +02001013 ret = actor(pipe, sd);
Tom Zanussia82c53a2008-05-09 13:28:36 +02001014 if (unlikely(ret <= 0)) {
1015 sd->pos = prev_pos;
Jens Axboeb92ce552006-04-11 13:52:07 +02001016 goto out_release;
Tom Zanussia82c53a2008-05-09 13:28:36 +02001017 }
Jens Axboeb92ce552006-04-11 13:52:07 +02001018
1019 bytes += ret;
1020 len -= ret;
Jens Axboebcd4f3a2007-07-16 14:41:49 +02001021 sd->pos = pos;
Jens Axboeb92ce552006-04-11 13:52:07 +02001022
Tom Zanussia82c53a2008-05-09 13:28:36 +02001023 if (ret < read_len) {
1024 sd->pos = prev_pos + ret;
Jens Axboe51a92c02007-07-13 14:11:43 +02001025 goto out_release;
Tom Zanussia82c53a2008-05-09 13:28:36 +02001026 }
Jens Axboeb92ce552006-04-11 13:52:07 +02001027 }
1028
Jens Axboe9e971982008-01-29 21:05:57 +01001029done:
Jens Axboeb92ce552006-04-11 13:52:07 +02001030 pipe->nrbufs = pipe->curbuf = 0;
Jens Axboe80848702008-01-30 12:24:48 +01001031 file_accessed(in);
Jens Axboeb92ce552006-04-11 13:52:07 +02001032 return bytes;
1033
1034out_release:
1035 /*
1036 * If we did an incomplete transfer we must release
1037 * the pipe buffers in question:
1038 */
1039 for (i = 0; i < PIPE_BUFFERS; i++) {
1040 struct pipe_buffer *buf = pipe->bufs + i;
1041
1042 if (buf->ops) {
1043 buf->ops->release(pipe, buf);
1044 buf->ops = NULL;
1045 }
1046 }
Jens Axboeb92ce552006-04-11 13:52:07 +02001047
Jens Axboe9e971982008-01-29 21:05:57 +01001048 if (!bytes)
1049 bytes = ret;
Jens Axboeb92ce552006-04-11 13:52:07 +02001050
Jens Axboe9e971982008-01-29 21:05:57 +01001051 goto done;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001052}
1053EXPORT_SYMBOL(splice_direct_to_actor);
1054
1055static int direct_splice_actor(struct pipe_inode_info *pipe,
1056 struct splice_desc *sd)
1057{
Jens Axboe6a14b902007-06-14 13:08:55 +02001058 struct file *file = sd->u.file;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001059
1060 return do_splice_from(pipe, file, &sd->pos, sd->total_len, sd->flags);
1061}
1062
Jens Axboe932cc6d2007-06-21 13:10:21 +02001063/**
1064 * do_splice_direct - splices data directly between two files
1065 * @in: file to splice from
1066 * @ppos: input file offset
1067 * @out: file to splice to
1068 * @len: number of bytes to splice
1069 * @flags: splice modifier flags
1070 *
1071 * Description:
1072 * For use by do_sendfile(). splice can easily emulate sendfile, but
1073 * doing it in the application would incur an extra system call
1074 * (splice in + splice out, as compared to just sendfile()). So this helper
1075 * can splice directly through a process-private pipe.
1076 *
1077 */
Jens Axboec66ab6f2007-06-12 21:17:17 +02001078long do_splice_direct(struct file *in, loff_t *ppos, struct file *out,
1079 size_t len, unsigned int flags)
1080{
1081 struct splice_desc sd = {
1082 .len = len,
1083 .total_len = len,
1084 .flags = flags,
1085 .pos = *ppos,
Jens Axboe6a14b902007-06-14 13:08:55 +02001086 .u.file = out,
Jens Axboec66ab6f2007-06-12 21:17:17 +02001087 };
Jens Axboe51a92c02007-07-13 14:11:43 +02001088 long ret;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001089
1090 ret = splice_direct_to_actor(in, &sd, direct_splice_actor);
Jens Axboe51a92c02007-07-13 14:11:43 +02001091 if (ret > 0)
Tom Zanussia82c53a2008-05-09 13:28:36 +02001092 *ppos = sd.pos;
Jens Axboe51a92c02007-07-13 14:11:43 +02001093
Jens Axboec66ab6f2007-06-12 21:17:17 +02001094 return ret;
Jens Axboeb92ce552006-04-11 13:52:07 +02001095}
1096
Jens Axboe83f91352006-04-02 23:05:09 +02001097/*
Jens Axboeddac0d32006-11-04 12:49:32 +01001098 * After the inode slimming patch, i_pipe/i_bdev/i_cdev share the same
1099 * location, so checking ->i_pipe is not enough to verify that this is a
1100 * pipe.
1101 */
1102static inline struct pipe_inode_info *pipe_info(struct inode *inode)
1103{
1104 if (S_ISFIFO(inode->i_mode))
1105 return inode->i_pipe;
1106
1107 return NULL;
1108}
1109
1110/*
Jens Axboe83f91352006-04-02 23:05:09 +02001111 * Determine where to splice to/from.
1112 */
Ingo Molnar529565d2006-04-10 15:18:58 +02001113static long do_splice(struct file *in, loff_t __user *off_in,
1114 struct file *out, loff_t __user *off_out,
1115 size_t len, unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +02001116{
Ingo Molnar3a326a22006-04-10 15:18:35 +02001117 struct pipe_inode_info *pipe;
Jens Axboecbb7e572006-04-11 14:57:50 +02001118 loff_t offset, *off;
Jens Axboea4514eb2006-04-19 15:57:05 +02001119 long ret;
Jens Axboe5274f052006-03-30 15:15:30 +02001120
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001121 pipe = pipe_info(in->f_path.dentry->d_inode);
Ingo Molnar529565d2006-04-10 15:18:58 +02001122 if (pipe) {
1123 if (off_in)
1124 return -ESPIPE;
Jens Axboeb92ce552006-04-11 13:52:07 +02001125 if (off_out) {
1126 if (out->f_op->llseek == no_llseek)
1127 return -EINVAL;
Jens Axboecbb7e572006-04-11 14:57:50 +02001128 if (copy_from_user(&offset, off_out, sizeof(loff_t)))
Jens Axboeb92ce552006-04-11 13:52:07 +02001129 return -EFAULT;
Jens Axboecbb7e572006-04-11 14:57:50 +02001130 off = &offset;
1131 } else
1132 off = &out->f_pos;
Ingo Molnar529565d2006-04-10 15:18:58 +02001133
Jens Axboea4514eb2006-04-19 15:57:05 +02001134 ret = do_splice_from(pipe, out, off, len, flags);
1135
1136 if (off_out && copy_to_user(off_out, off, sizeof(loff_t)))
1137 ret = -EFAULT;
1138
1139 return ret;
Ingo Molnar529565d2006-04-10 15:18:58 +02001140 }
Jens Axboe5274f052006-03-30 15:15:30 +02001141
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001142 pipe = pipe_info(out->f_path.dentry->d_inode);
Ingo Molnar529565d2006-04-10 15:18:58 +02001143 if (pipe) {
1144 if (off_out)
1145 return -ESPIPE;
Jens Axboeb92ce552006-04-11 13:52:07 +02001146 if (off_in) {
1147 if (in->f_op->llseek == no_llseek)
1148 return -EINVAL;
Jens Axboecbb7e572006-04-11 14:57:50 +02001149 if (copy_from_user(&offset, off_in, sizeof(loff_t)))
Jens Axboeb92ce552006-04-11 13:52:07 +02001150 return -EFAULT;
Jens Axboecbb7e572006-04-11 14:57:50 +02001151 off = &offset;
1152 } else
1153 off = &in->f_pos;
Ingo Molnar529565d2006-04-10 15:18:58 +02001154
Jens Axboea4514eb2006-04-19 15:57:05 +02001155 ret = do_splice_to(in, off, pipe, len, flags);
1156
1157 if (off_in && copy_to_user(off_in, off, sizeof(loff_t)))
1158 ret = -EFAULT;
1159
1160 return ret;
Ingo Molnar529565d2006-04-10 15:18:58 +02001161 }
Jens Axboe5274f052006-03-30 15:15:30 +02001162
1163 return -EINVAL;
1164}
1165
Jens Axboe912d35f2006-04-26 10:59:21 +02001166/*
1167 * Map an iov into an array of pages and offset/length tupples. With the
1168 * partial_page structure, we can map several non-contiguous ranges into
1169 * our ones pages[] map instead of splitting that operation into pieces.
1170 * Could easily be exported as a generic helper for other users, in which
1171 * case one would probably want to add a 'max_nr_pages' parameter as well.
1172 */
1173static int get_iovec_page_array(const struct iovec __user *iov,
1174 unsigned int nr_vecs, struct page **pages,
Jens Axboe7afa6fd2006-05-01 20:02:33 +02001175 struct partial_page *partial, int aligned)
Jens Axboe912d35f2006-04-26 10:59:21 +02001176{
1177 int buffers = 0, error = 0;
1178
Jens Axboe912d35f2006-04-26 10:59:21 +02001179 while (nr_vecs) {
1180 unsigned long off, npages;
Linus Torvalds75723952007-10-01 13:17:28 -07001181 struct iovec entry;
Jens Axboe912d35f2006-04-26 10:59:21 +02001182 void __user *base;
1183 size_t len;
1184 int i;
1185
Linus Torvalds75723952007-10-01 13:17:28 -07001186 error = -EFAULT;
Nick Pigginbc40d732008-07-25 19:45:26 -07001187 if (copy_from_user(&entry, iov, sizeof(entry)))
Jens Axboe912d35f2006-04-26 10:59:21 +02001188 break;
Linus Torvalds75723952007-10-01 13:17:28 -07001189
1190 base = entry.iov_base;
1191 len = entry.iov_len;
Jens Axboe912d35f2006-04-26 10:59:21 +02001192
1193 /*
1194 * Sanity check this iovec. 0 read succeeds.
1195 */
Linus Torvalds75723952007-10-01 13:17:28 -07001196 error = 0;
Jens Axboe912d35f2006-04-26 10:59:21 +02001197 if (unlikely(!len))
1198 break;
1199 error = -EFAULT;
Bastian Blank712a30e2008-02-10 16:47:57 +02001200 if (!access_ok(VERIFY_READ, base, len))
Jens Axboe912d35f2006-04-26 10:59:21 +02001201 break;
1202
1203 /*
1204 * Get this base offset and number of pages, then map
1205 * in the user pages.
1206 */
1207 off = (unsigned long) base & ~PAGE_MASK;
Jens Axboe7afa6fd2006-05-01 20:02:33 +02001208
1209 /*
1210 * If asked for alignment, the offset must be zero and the
1211 * length a multiple of the PAGE_SIZE.
1212 */
1213 error = -EINVAL;
1214 if (aligned && (off || len & ~PAGE_MASK))
1215 break;
1216
Jens Axboe912d35f2006-04-26 10:59:21 +02001217 npages = (off + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1218 if (npages > PIPE_BUFFERS - buffers)
1219 npages = PIPE_BUFFERS - buffers;
1220
Nick Pigginbc40d732008-07-25 19:45:26 -07001221 error = get_user_pages_fast((unsigned long)base, npages,
1222 0, &pages[buffers]);
Jens Axboe912d35f2006-04-26 10:59:21 +02001223
1224 if (unlikely(error <= 0))
1225 break;
1226
1227 /*
1228 * Fill this contiguous range into the partial page map.
1229 */
1230 for (i = 0; i < error; i++) {
Jens Axboe75914892006-05-02 12:57:18 +02001231 const int plen = min_t(size_t, len, PAGE_SIZE - off);
Jens Axboe912d35f2006-04-26 10:59:21 +02001232
1233 partial[buffers].offset = off;
1234 partial[buffers].len = plen;
1235
1236 off = 0;
1237 len -= plen;
1238 buffers++;
1239 }
1240
1241 /*
1242 * We didn't complete this iov, stop here since it probably
1243 * means we have to move some of this into a pipe to
1244 * be able to continue.
1245 */
1246 if (len)
1247 break;
1248
1249 /*
1250 * Don't continue if we mapped fewer pages than we asked for,
1251 * or if we mapped the max number of pages that we have
1252 * room for.
1253 */
1254 if (error < npages || buffers == PIPE_BUFFERS)
1255 break;
1256
1257 nr_vecs--;
1258 iov++;
1259 }
1260
Jens Axboe912d35f2006-04-26 10:59:21 +02001261 if (buffers)
1262 return buffers;
1263
1264 return error;
1265}
1266
Jens Axboe6a14b902007-06-14 13:08:55 +02001267static int pipe_to_user(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
1268 struct splice_desc *sd)
1269{
1270 char *src;
1271 int ret;
1272
Jens Axboecac36bb02007-06-14 13:10:48 +02001273 ret = buf->ops->confirm(pipe, buf);
Jens Axboe6a14b902007-06-14 13:08:55 +02001274 if (unlikely(ret))
1275 return ret;
1276
1277 /*
1278 * See if we can use the atomic maps, by prefaulting in the
1279 * pages and doing an atomic copy
1280 */
1281 if (!fault_in_pages_writeable(sd->u.userptr, sd->len)) {
1282 src = buf->ops->map(pipe, buf, 1);
1283 ret = __copy_to_user_inatomic(sd->u.userptr, src + buf->offset,
1284 sd->len);
1285 buf->ops->unmap(pipe, buf, src);
1286 if (!ret) {
1287 ret = sd->len;
1288 goto out;
1289 }
1290 }
1291
1292 /*
1293 * No dice, use slow non-atomic map and copy
1294 */
1295 src = buf->ops->map(pipe, buf, 0);
1296
1297 ret = sd->len;
1298 if (copy_to_user(sd->u.userptr, src + buf->offset, sd->len))
1299 ret = -EFAULT;
1300
Jens Axboe6866bef2007-10-16 10:01:29 +02001301 buf->ops->unmap(pipe, buf, src);
Jens Axboe6a14b902007-06-14 13:08:55 +02001302out:
1303 if (ret > 0)
1304 sd->u.userptr += ret;
Jens Axboe6a14b902007-06-14 13:08:55 +02001305 return ret;
1306}
1307
1308/*
1309 * For lack of a better implementation, implement vmsplice() to userspace
1310 * as a simple copy of the pipes pages to the user iov.
1311 */
1312static long vmsplice_to_user(struct file *file, const struct iovec __user *iov,
1313 unsigned long nr_segs, unsigned int flags)
1314{
1315 struct pipe_inode_info *pipe;
1316 struct splice_desc sd;
1317 ssize_t size;
1318 int error;
1319 long ret;
1320
1321 pipe = pipe_info(file->f_path.dentry->d_inode);
1322 if (!pipe)
1323 return -EBADF;
1324
1325 if (pipe->inode)
1326 mutex_lock(&pipe->inode->i_mutex);
1327
1328 error = ret = 0;
1329 while (nr_segs) {
1330 void __user *base;
1331 size_t len;
1332
1333 /*
1334 * Get user address base and length for this iovec.
1335 */
1336 error = get_user(base, &iov->iov_base);
1337 if (unlikely(error))
1338 break;
1339 error = get_user(len, &iov->iov_len);
1340 if (unlikely(error))
1341 break;
1342
1343 /*
1344 * Sanity check this iovec. 0 read succeeds.
1345 */
1346 if (unlikely(!len))
1347 break;
1348 if (unlikely(!base)) {
1349 error = -EFAULT;
1350 break;
1351 }
1352
Jens Axboe88119302008-02-08 08:49:14 -08001353 if (unlikely(!access_ok(VERIFY_WRITE, base, len))) {
1354 error = -EFAULT;
1355 break;
1356 }
1357
Jens Axboe6a14b902007-06-14 13:08:55 +02001358 sd.len = 0;
1359 sd.total_len = len;
1360 sd.flags = flags;
1361 sd.u.userptr = base;
1362 sd.pos = 0;
1363
1364 size = __splice_from_pipe(pipe, &sd, pipe_to_user);
1365 if (size < 0) {
1366 if (!ret)
1367 ret = size;
1368
1369 break;
1370 }
1371
1372 ret += size;
1373
1374 if (size < len)
1375 break;
1376
1377 nr_segs--;
1378 iov++;
1379 }
1380
1381 if (pipe->inode)
1382 mutex_unlock(&pipe->inode->i_mutex);
1383
1384 if (!ret)
1385 ret = error;
1386
1387 return ret;
1388}
1389
Jens Axboe912d35f2006-04-26 10:59:21 +02001390/*
1391 * vmsplice splices a user address range into a pipe. It can be thought of
1392 * as splice-from-memory, where the regular splice is splice-from-file (or
1393 * to file). In both cases the output is a pipe, naturally.
Jens Axboe912d35f2006-04-26 10:59:21 +02001394 */
Jens Axboe6a14b902007-06-14 13:08:55 +02001395static long vmsplice_to_pipe(struct file *file, const struct iovec __user *iov,
1396 unsigned long nr_segs, unsigned int flags)
Jens Axboe912d35f2006-04-26 10:59:21 +02001397{
Jens Axboeddac0d32006-11-04 12:49:32 +01001398 struct pipe_inode_info *pipe;
Jens Axboe912d35f2006-04-26 10:59:21 +02001399 struct page *pages[PIPE_BUFFERS];
1400 struct partial_page partial[PIPE_BUFFERS];
1401 struct splice_pipe_desc spd = {
1402 .pages = pages,
1403 .partial = partial,
1404 .flags = flags,
1405 .ops = &user_page_pipe_buf_ops,
Jens Axboebbdfc2f2007-11-06 23:29:47 -08001406 .spd_release = spd_release_page,
Jens Axboe912d35f2006-04-26 10:59:21 +02001407 };
1408
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001409 pipe = pipe_info(file->f_path.dentry->d_inode);
Jens Axboeddac0d32006-11-04 12:49:32 +01001410 if (!pipe)
Jens Axboe912d35f2006-04-26 10:59:21 +02001411 return -EBADF;
Jens Axboe912d35f2006-04-26 10:59:21 +02001412
Jens Axboe7afa6fd2006-05-01 20:02:33 +02001413 spd.nr_pages = get_iovec_page_array(iov, nr_segs, pages, partial,
1414 flags & SPLICE_F_GIFT);
Jens Axboe912d35f2006-04-26 10:59:21 +02001415 if (spd.nr_pages <= 0)
1416 return spd.nr_pages;
1417
Jens Axboe00522fb2006-04-26 14:39:29 +02001418 return splice_to_pipe(pipe, &spd);
Jens Axboe912d35f2006-04-26 10:59:21 +02001419}
1420
Jens Axboe6a14b902007-06-14 13:08:55 +02001421/*
1422 * Note that vmsplice only really supports true splicing _from_ user memory
1423 * to a pipe, not the other way around. Splicing from user memory is a simple
1424 * operation that can be supported without any funky alignment restrictions
1425 * or nasty vm tricks. We simply map in the user memory and fill them into
1426 * a pipe. The reverse isn't quite as easy, though. There are two possible
1427 * solutions for that:
1428 *
1429 * - memcpy() the data internally, at which point we might as well just
1430 * do a regular read() on the buffer anyway.
1431 * - Lots of nasty vm tricks, that are neither fast nor flexible (it
1432 * has restriction limitations on both ends of the pipe).
1433 *
1434 * Currently we punt and implement it as a normal copy, see pipe_to_user().
1435 *
1436 */
Jens Axboe912d35f2006-04-26 10:59:21 +02001437asmlinkage long sys_vmsplice(int fd, const struct iovec __user *iov,
1438 unsigned long nr_segs, unsigned int flags)
1439{
1440 struct file *file;
1441 long error;
1442 int fput;
1443
Jens Axboe6a14b902007-06-14 13:08:55 +02001444 if (unlikely(nr_segs > UIO_MAXIOV))
1445 return -EINVAL;
1446 else if (unlikely(!nr_segs))
1447 return 0;
1448
Jens Axboe912d35f2006-04-26 10:59:21 +02001449 error = -EBADF;
1450 file = fget_light(fd, &fput);
1451 if (file) {
1452 if (file->f_mode & FMODE_WRITE)
Jens Axboe6a14b902007-06-14 13:08:55 +02001453 error = vmsplice_to_pipe(file, iov, nr_segs, flags);
1454 else if (file->f_mode & FMODE_READ)
1455 error = vmsplice_to_user(file, iov, nr_segs, flags);
Jens Axboe912d35f2006-04-26 10:59:21 +02001456
1457 fput_light(file, fput);
1458 }
1459
1460 return error;
1461}
1462
Ingo Molnar529565d2006-04-10 15:18:58 +02001463asmlinkage long sys_splice(int fd_in, loff_t __user *off_in,
1464 int fd_out, loff_t __user *off_out,
1465 size_t len, unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +02001466{
1467 long error;
1468 struct file *in, *out;
1469 int fput_in, fput_out;
1470
1471 if (unlikely(!len))
1472 return 0;
1473
1474 error = -EBADF;
Ingo Molnar529565d2006-04-10 15:18:58 +02001475 in = fget_light(fd_in, &fput_in);
Jens Axboe5274f052006-03-30 15:15:30 +02001476 if (in) {
1477 if (in->f_mode & FMODE_READ) {
Ingo Molnar529565d2006-04-10 15:18:58 +02001478 out = fget_light(fd_out, &fput_out);
Jens Axboe5274f052006-03-30 15:15:30 +02001479 if (out) {
1480 if (out->f_mode & FMODE_WRITE)
Ingo Molnar529565d2006-04-10 15:18:58 +02001481 error = do_splice(in, off_in,
1482 out, off_out,
1483 len, flags);
Jens Axboe5274f052006-03-30 15:15:30 +02001484 fput_light(out, fput_out);
1485 }
1486 }
1487
1488 fput_light(in, fput_in);
1489 }
1490
1491 return error;
1492}
Jens Axboe70524492006-04-11 15:51:17 +02001493
1494/*
Jens Axboeaadd06e2006-07-10 11:00:01 +02001495 * Make sure there's data to read. Wait for input if we can, otherwise
1496 * return an appropriate error.
1497 */
1498static int link_ipipe_prep(struct pipe_inode_info *pipe, unsigned int flags)
1499{
1500 int ret;
1501
1502 /*
1503 * Check ->nrbufs without the inode lock first. This function
1504 * is speculative anyways, so missing one is ok.
1505 */
1506 if (pipe->nrbufs)
1507 return 0;
1508
1509 ret = 0;
1510 mutex_lock(&pipe->inode->i_mutex);
1511
1512 while (!pipe->nrbufs) {
1513 if (signal_pending(current)) {
1514 ret = -ERESTARTSYS;
1515 break;
1516 }
1517 if (!pipe->writers)
1518 break;
1519 if (!pipe->waiting_writers) {
1520 if (flags & SPLICE_F_NONBLOCK) {
1521 ret = -EAGAIN;
1522 break;
1523 }
1524 }
1525 pipe_wait(pipe);
1526 }
1527
1528 mutex_unlock(&pipe->inode->i_mutex);
1529 return ret;
1530}
1531
1532/*
1533 * Make sure there's writeable room. Wait for room if we can, otherwise
1534 * return an appropriate error.
1535 */
1536static int link_opipe_prep(struct pipe_inode_info *pipe, unsigned int flags)
1537{
1538 int ret;
1539
1540 /*
1541 * Check ->nrbufs without the inode lock first. This function
1542 * is speculative anyways, so missing one is ok.
1543 */
1544 if (pipe->nrbufs < PIPE_BUFFERS)
1545 return 0;
1546
1547 ret = 0;
1548 mutex_lock(&pipe->inode->i_mutex);
1549
1550 while (pipe->nrbufs >= PIPE_BUFFERS) {
1551 if (!pipe->readers) {
1552 send_sig(SIGPIPE, current, 0);
1553 ret = -EPIPE;
1554 break;
1555 }
1556 if (flags & SPLICE_F_NONBLOCK) {
1557 ret = -EAGAIN;
1558 break;
1559 }
1560 if (signal_pending(current)) {
1561 ret = -ERESTARTSYS;
1562 break;
1563 }
1564 pipe->waiting_writers++;
1565 pipe_wait(pipe);
1566 pipe->waiting_writers--;
1567 }
1568
1569 mutex_unlock(&pipe->inode->i_mutex);
1570 return ret;
1571}
1572
1573/*
Jens Axboe70524492006-04-11 15:51:17 +02001574 * Link contents of ipipe to opipe.
1575 */
1576static int link_pipe(struct pipe_inode_info *ipipe,
1577 struct pipe_inode_info *opipe,
1578 size_t len, unsigned int flags)
1579{
1580 struct pipe_buffer *ibuf, *obuf;
Jens Axboeaadd06e2006-07-10 11:00:01 +02001581 int ret = 0, i = 0, nbuf;
Jens Axboe70524492006-04-11 15:51:17 +02001582
1583 /*
1584 * Potential ABBA deadlock, work around it by ordering lock
1585 * grabbing by inode address. Otherwise two different processes
1586 * could deadlock (one doing tee from A -> B, the other from B -> A).
1587 */
Mark Fasheh62752ee2006-10-17 10:31:38 +02001588 inode_double_lock(ipipe->inode, opipe->inode);
Jens Axboe70524492006-04-11 15:51:17 +02001589
Jens Axboeaadd06e2006-07-10 11:00:01 +02001590 do {
Jens Axboe70524492006-04-11 15:51:17 +02001591 if (!opipe->readers) {
1592 send_sig(SIGPIPE, current, 0);
1593 if (!ret)
1594 ret = -EPIPE;
1595 break;
1596 }
Jens Axboe70524492006-04-11 15:51:17 +02001597
1598 /*
Jens Axboeaadd06e2006-07-10 11:00:01 +02001599 * If we have iterated all input buffers or ran out of
1600 * output room, break.
Jens Axboe70524492006-04-11 15:51:17 +02001601 */
Jens Axboeaadd06e2006-07-10 11:00:01 +02001602 if (i >= ipipe->nrbufs || opipe->nrbufs >= PIPE_BUFFERS)
Jens Axboe70524492006-04-11 15:51:17 +02001603 break;
Jens Axboeaadd06e2006-07-10 11:00:01 +02001604
1605 ibuf = ipipe->bufs + ((ipipe->curbuf + i) & (PIPE_BUFFERS - 1));
1606 nbuf = (opipe->curbuf + opipe->nrbufs) & (PIPE_BUFFERS - 1);
1607
Jens Axboe2a27250e2006-04-19 15:56:40 +02001608 /*
Jens Axboeaadd06e2006-07-10 11:00:01 +02001609 * Get a reference to this pipe buffer,
1610 * so we can copy the contents over.
Jens Axboe2a27250e2006-04-19 15:56:40 +02001611 */
Jens Axboeaadd06e2006-07-10 11:00:01 +02001612 ibuf->ops->get(ipipe, ibuf);
Jens Axboe70524492006-04-11 15:51:17 +02001613
Jens Axboeaadd06e2006-07-10 11:00:01 +02001614 obuf = opipe->bufs + nbuf;
1615 *obuf = *ibuf;
Jens Axboe70524492006-04-11 15:51:17 +02001616
Jens Axboeaadd06e2006-07-10 11:00:01 +02001617 /*
1618 * Don't inherit the gift flag, we need to
1619 * prevent multiple steals of this page.
1620 */
1621 obuf->flags &= ~PIPE_BUF_FLAG_GIFT;
1622
1623 if (obuf->len > len)
1624 obuf->len = len;
1625
1626 opipe->nrbufs++;
1627 ret += obuf->len;
1628 len -= obuf->len;
1629 i++;
1630 } while (len);
Jens Axboe70524492006-04-11 15:51:17 +02001631
Jens Axboe02cf01a2008-02-20 10:34:51 +01001632 /*
1633 * return EAGAIN if we have the potential of some data in the
1634 * future, otherwise just return 0
1635 */
1636 if (!ret && ipipe->waiting_writers && (flags & SPLICE_F_NONBLOCK))
1637 ret = -EAGAIN;
1638
Mark Fasheh62752ee2006-10-17 10:31:38 +02001639 inode_double_unlock(ipipe->inode, opipe->inode);
Jens Axboe70524492006-04-11 15:51:17 +02001640
Jens Axboeaadd06e2006-07-10 11:00:01 +02001641 /*
1642 * If we put data in the output pipe, wakeup any potential readers.
1643 */
1644 if (ret > 0) {
Jens Axboe70524492006-04-11 15:51:17 +02001645 smp_mb();
1646 if (waitqueue_active(&opipe->wait))
1647 wake_up_interruptible(&opipe->wait);
1648 kill_fasync(&opipe->fasync_readers, SIGIO, POLL_IN);
1649 }
1650
1651 return ret;
1652}
1653
1654/*
1655 * This is a tee(1) implementation that works on pipes. It doesn't copy
1656 * any data, it simply references the 'in' pages on the 'out' pipe.
1657 * The 'flags' used are the SPLICE_F_* variants, currently the only
1658 * applicable one is SPLICE_F_NONBLOCK.
1659 */
1660static long do_tee(struct file *in, struct file *out, size_t len,
1661 unsigned int flags)
1662{
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001663 struct pipe_inode_info *ipipe = pipe_info(in->f_path.dentry->d_inode);
1664 struct pipe_inode_info *opipe = pipe_info(out->f_path.dentry->d_inode);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001665 int ret = -EINVAL;
Jens Axboe70524492006-04-11 15:51:17 +02001666
1667 /*
Jens Axboeaadd06e2006-07-10 11:00:01 +02001668 * Duplicate the contents of ipipe to opipe without actually
1669 * copying the data.
Jens Axboe70524492006-04-11 15:51:17 +02001670 */
Jens Axboeaadd06e2006-07-10 11:00:01 +02001671 if (ipipe && opipe && ipipe != opipe) {
1672 /*
1673 * Keep going, unless we encounter an error. The ipipe/opipe
1674 * ordering doesn't really matter.
1675 */
1676 ret = link_ipipe_prep(ipipe, flags);
1677 if (!ret) {
1678 ret = link_opipe_prep(opipe, flags);
Jens Axboe02cf01a2008-02-20 10:34:51 +01001679 if (!ret)
Jens Axboeaadd06e2006-07-10 11:00:01 +02001680 ret = link_pipe(ipipe, opipe, len, flags);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001681 }
1682 }
Jens Axboe70524492006-04-11 15:51:17 +02001683
Jens Axboeaadd06e2006-07-10 11:00:01 +02001684 return ret;
Jens Axboe70524492006-04-11 15:51:17 +02001685}
1686
1687asmlinkage long sys_tee(int fdin, int fdout, size_t len, unsigned int flags)
1688{
1689 struct file *in;
1690 int error, fput_in;
1691
1692 if (unlikely(!len))
1693 return 0;
1694
1695 error = -EBADF;
1696 in = fget_light(fdin, &fput_in);
1697 if (in) {
1698 if (in->f_mode & FMODE_READ) {
1699 int fput_out;
1700 struct file *out = fget_light(fdout, &fput_out);
1701
1702 if (out) {
1703 if (out->f_mode & FMODE_WRITE)
1704 error = do_tee(in, out, len, flags);
1705 fput_light(out, fput_out);
1706 }
1707 }
1708 fput_light(in, fput_in);
1709 }
1710
1711 return error;
1712}