blob: 59a941d404d9c60b46a83ced4472ee915a7fe93c [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 Axboe912d35f2006-04-26 10:59:21 +0200257 page_cache_release(spd->pages[page_nr++]);
Jens Axboe5274f052006-03-30 15:15:30 +0200258
259 return ret;
260}
261
Ingo Molnar3a326a22006-04-10 15:18:35 +0200262static int
Jens Axboecbb7e572006-04-11 14:57:50 +0200263__generic_file_splice_read(struct file *in, loff_t *ppos,
264 struct pipe_inode_info *pipe, size_t len,
265 unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +0200266{
267 struct address_space *mapping = in->f_mapping;
Fengguang Wud8983912007-07-19 01:48:06 -0700268 unsigned int loff, nr_pages, req_pages;
Jens Axboe16c523d2006-04-10 09:03:58 +0200269 struct page *pages[PIPE_BUFFERS];
Jens Axboe912d35f2006-04-26 10:59:21 +0200270 struct partial_page partial[PIPE_BUFFERS];
Jens Axboe5274f052006-03-30 15:15:30 +0200271 struct page *page;
Jens Axboe91ad66e2006-04-19 15:55:10 +0200272 pgoff_t index, end_index;
273 loff_t isize;
Jens Axboeeb207962006-04-27 11:05:22 +0200274 int error, page_nr;
Jens Axboe912d35f2006-04-26 10:59:21 +0200275 struct splice_pipe_desc spd = {
276 .pages = pages,
277 .partial = partial,
278 .flags = flags,
279 .ops = &page_cache_pipe_buf_ops,
280 };
Jens Axboe5274f052006-03-30 15:15:30 +0200281
Jens Axboecbb7e572006-04-11 14:57:50 +0200282 index = *ppos >> PAGE_CACHE_SHIFT;
Jens Axboe912d35f2006-04-26 10:59:21 +0200283 loff = *ppos & ~PAGE_CACHE_MASK;
Fengguang Wud8983912007-07-19 01:48:06 -0700284 req_pages = (len + loff + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
285 nr_pages = min(req_pages, (unsigned)PIPE_BUFFERS);
Jens Axboe5274f052006-03-30 15:15:30 +0200286
287 /*
Jens Axboeeb207962006-04-27 11:05:22 +0200288 * Lookup the (hopefully) full range of pages we need.
289 */
290 spd.nr_pages = find_get_pages_contig(mapping, index, nr_pages, pages);
Fengguang Wu431a4822007-07-19 01:48:05 -0700291 index += spd.nr_pages;
Jens Axboeeb207962006-04-27 11:05:22 +0200292
293 /*
294 * If find_get_pages_contig() returned fewer pages than we needed,
Fengguang Wu431a4822007-07-19 01:48:05 -0700295 * readahead/allocate the rest and fill in the holes.
Jens Axboeeb207962006-04-27 11:05:22 +0200296 */
Fengguang Wu431a4822007-07-19 01:48:05 -0700297 if (spd.nr_pages < nr_pages)
Rusty Russellcf914a72007-07-19 01:48:08 -0700298 page_cache_sync_readahead(mapping, &in->f_ra, in,
299 index, req_pages - spd.nr_pages);
Fengguang Wu431a4822007-07-19 01:48:05 -0700300
Jens Axboe932cc6d2007-06-21 13:10:21 +0200301 error = 0;
Jens Axboeeb207962006-04-27 11:05:22 +0200302 while (spd.nr_pages < nr_pages) {
303 /*
304 * Page could be there, find_get_pages_contig() breaks on
305 * the first hole.
306 */
307 page = find_get_page(mapping, index);
308 if (!page) {
Jens Axboee27dedd2006-05-01 19:59:54 +0200309 /*
Jens Axboeeb207962006-04-27 11:05:22 +0200310 * page didn't exist, allocate one.
311 */
312 page = page_cache_alloc_cold(mapping);
313 if (!page)
314 break;
315
316 error = add_to_page_cache_lru(page, mapping, index,
Nick Piggin2ae88142006-10-28 10:38:23 -0700317 GFP_KERNEL);
Jens Axboeeb207962006-04-27 11:05:22 +0200318 if (unlikely(error)) {
319 page_cache_release(page);
Jens Axboea0548872006-05-03 10:58:22 +0200320 if (error == -EEXIST)
321 continue;
Jens Axboeeb207962006-04-27 11:05:22 +0200322 break;
323 }
324 /*
325 * add_to_page_cache() locks the page, unlock it
326 * to avoid convoluting the logic below even more.
327 */
328 unlock_page(page);
329 }
330
331 pages[spd.nr_pages++] = page;
332 index++;
333 }
334
335 /*
336 * Now loop over the map and see if we need to start IO on any
337 * pages, fill in the partial map, etc.
338 */
339 index = *ppos >> PAGE_CACHE_SHIFT;
340 nr_pages = spd.nr_pages;
341 spd.nr_pages = 0;
342 for (page_nr = 0; page_nr < nr_pages; page_nr++) {
Jens Axboe82aa5d62006-04-20 13:05:48 +0200343 unsigned int this_len;
344
345 if (!len)
346 break;
347
348 /*
349 * this_len is the max we'll use from this page
350 */
Andrew Mortonba5f5d92006-04-25 15:33:34 +0200351 this_len = min_t(unsigned long, len, PAGE_CACHE_SIZE - loff);
Jens Axboeeb207962006-04-27 11:05:22 +0200352 page = pages[page_nr];
Jens Axboe7480a902006-04-11 13:52:47 +0200353
Fengguang Wua08a1662007-07-19 01:48:03 -0700354 if (PageReadahead(page))
Rusty Russellcf914a72007-07-19 01:48:08 -0700355 page_cache_async_readahead(mapping, &in->f_ra, in,
Fengguang Wud8983912007-07-19 01:48:06 -0700356 page, index, req_pages - page_nr);
Fengguang Wua08a1662007-07-19 01:48:03 -0700357
Jens Axboe7480a902006-04-11 13:52:47 +0200358 /*
359 * If the page isn't uptodate, we may need to start io on it
360 */
361 if (!PageUptodate(page)) {
Jens Axboec4f895c2006-04-19 15:56:12 +0200362 /*
363 * If in nonblock mode then dont block on waiting
364 * for an in-flight io page
365 */
Fengguang Wu9ae9d682007-05-08 08:44:36 +0200366 if (flags & SPLICE_F_NONBLOCK) {
367 if (TestSetPageLocked(page))
368 break;
369 } else
370 lock_page(page);
Jens Axboe7480a902006-04-11 13:52:47 +0200371
372 /*
373 * page was truncated, stop here. if this isn't the
374 * first page, we'll just complete what we already
375 * added
376 */
377 if (!page->mapping) {
378 unlock_page(page);
Jens Axboe7480a902006-04-11 13:52:47 +0200379 break;
380 }
381 /*
382 * page was already under io and is now done, great
383 */
384 if (PageUptodate(page)) {
385 unlock_page(page);
386 goto fill_it;
387 }
388
Jens Axboe7480a902006-04-11 13:52:47 +0200389 /*
390 * need to read in the page
391 */
392 error = mapping->a_ops->readpage(in, page);
Jens Axboe7480a902006-04-11 13:52:47 +0200393 if (unlikely(error)) {
Jens Axboeeb207962006-04-27 11:05:22 +0200394 /*
395 * We really should re-lookup the page here,
396 * but it complicates things a lot. Instead
397 * lets just do what we already stored, and
398 * we'll get it the next time we are called.
399 */
Jens Axboe7480a902006-04-11 13:52:47 +0200400 if (error == AOP_TRUNCATED_PAGE)
Jens Axboeeb207962006-04-27 11:05:22 +0200401 error = 0;
402
Jens Axboe7480a902006-04-11 13:52:47 +0200403 break;
404 }
Jens Axboe620a3242007-06-07 09:39:42 +0200405 }
406fill_it:
407 /*
408 * i_size must be checked after PageUptodate.
409 */
410 isize = i_size_read(mapping->host);
411 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
412 if (unlikely(!isize || index > end_index))
413 break;
414
415 /*
416 * if this is the last page, see if we need to shrink
417 * the length and stop
418 */
419 if (end_index == index) {
420 unsigned int plen;
Jens Axboe91ad66e2006-04-19 15:55:10 +0200421
422 /*
Jens Axboe620a3242007-06-07 09:39:42 +0200423 * max good bytes in this page
Jens Axboe91ad66e2006-04-19 15:55:10 +0200424 */
Jens Axboe620a3242007-06-07 09:39:42 +0200425 plen = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
426 if (plen <= loff)
Jens Axboe91ad66e2006-04-19 15:55:10 +0200427 break;
Jens Axboe91ad66e2006-04-19 15:55:10 +0200428
429 /*
Jens Axboe620a3242007-06-07 09:39:42 +0200430 * force quit after adding this page
Jens Axboe91ad66e2006-04-19 15:55:10 +0200431 */
Jens Axboe620a3242007-06-07 09:39:42 +0200432 this_len = min(this_len, plen - loff);
433 len = this_len;
Jens Axboe7480a902006-04-11 13:52:47 +0200434 }
Jens Axboe620a3242007-06-07 09:39:42 +0200435
Jens Axboeeb207962006-04-27 11:05:22 +0200436 partial[page_nr].offset = loff;
437 partial[page_nr].len = this_len;
Jens Axboe82aa5d62006-04-20 13:05:48 +0200438 len -= this_len;
Jens Axboe91ad66e2006-04-19 15:55:10 +0200439 loff = 0;
Jens Axboeeb207962006-04-27 11:05:22 +0200440 spd.nr_pages++;
441 index++;
Jens Axboe5274f052006-03-30 15:15:30 +0200442 }
443
Jens Axboeeb207962006-04-27 11:05:22 +0200444 /*
Hugh Dickins475ecad2007-06-07 09:36:00 +0200445 * Release any pages at the end, if we quit early. 'page_nr' is how far
Jens Axboeeb207962006-04-27 11:05:22 +0200446 * we got, 'nr_pages' is how many pages are in the map.
447 */
448 while (page_nr < nr_pages)
449 page_cache_release(pages[page_nr++]);
Fengguang Wuf4e6b492007-10-16 01:24:33 -0700450 in->f_ra.prev_pos = (loff_t)index << PAGE_CACHE_SHIFT;
Jens Axboeeb207962006-04-27 11:05:22 +0200451
Jens Axboe912d35f2006-04-26 10:59:21 +0200452 if (spd.nr_pages)
Jens Axboe00522fb2006-04-26 14:39:29 +0200453 return splice_to_pipe(pipe, &spd);
Jens Axboe5274f052006-03-30 15:15:30 +0200454
Jens Axboe7480a902006-04-11 13:52:47 +0200455 return error;
Jens Axboe5274f052006-03-30 15:15:30 +0200456}
457
Jens Axboe83f91352006-04-02 23:05:09 +0200458/**
459 * generic_file_splice_read - splice data from file to a pipe
460 * @in: file to splice from
Jens Axboe932cc6d2007-06-21 13:10:21 +0200461 * @ppos: position in @in
Jens Axboe83f91352006-04-02 23:05:09 +0200462 * @pipe: pipe to splice to
463 * @len: number of bytes to splice
464 * @flags: splice modifier flags
465 *
Jens Axboe932cc6d2007-06-21 13:10:21 +0200466 * Description:
467 * Will read pages from given file and fill them into a pipe. Can be
468 * used as long as the address_space operations for the source implements
469 * a readpage() hook.
470 *
Jens Axboe83f91352006-04-02 23:05:09 +0200471 */
Jens Axboecbb7e572006-04-11 14:57:50 +0200472ssize_t generic_file_splice_read(struct file *in, loff_t *ppos,
473 struct pipe_inode_info *pipe, size_t len,
474 unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +0200475{
476 ssize_t spliced;
477 int ret;
Jens Axboed366d3982007-06-01 14:54:11 +0200478 loff_t isize, left;
479
480 isize = i_size_read(in->f_mapping->host);
481 if (unlikely(*ppos >= isize))
482 return 0;
483
484 left = isize - *ppos;
485 if (unlikely(left < len))
486 len = left;
Jens Axboe5274f052006-03-30 15:15:30 +0200487
488 ret = 0;
489 spliced = 0;
Jens Axboe51a92c02007-07-13 14:11:43 +0200490 while (len && !spliced) {
Jens Axboecbb7e572006-04-11 14:57:50 +0200491 ret = __generic_file_splice_read(in, ppos, pipe, len, flags);
Jens Axboe5274f052006-03-30 15:15:30 +0200492
Jens Axboec4f895c2006-04-19 15:56:12 +0200493 if (ret < 0)
Jens Axboe5274f052006-03-30 15:15:30 +0200494 break;
Jens Axboec4f895c2006-04-19 15:56:12 +0200495 else if (!ret) {
496 if (spliced)
497 break;
498 if (flags & SPLICE_F_NONBLOCK) {
499 ret = -EAGAIN;
500 break;
501 }
502 }
Jens Axboe5274f052006-03-30 15:15:30 +0200503
Jens Axboecbb7e572006-04-11 14:57:50 +0200504 *ppos += ret;
Jens Axboe5274f052006-03-30 15:15:30 +0200505 len -= ret;
506 spliced += ret;
507 }
508
509 if (spliced)
510 return spliced;
511
512 return ret;
513}
514
Jens Axboe059a8f32006-04-02 23:06:05 +0200515EXPORT_SYMBOL(generic_file_splice_read);
516
Jens Axboe5274f052006-03-30 15:15:30 +0200517/*
Jens Axboe4f6f0bd2006-04-02 23:04:46 +0200518 * Send 'sd->len' bytes to socket from 'sd->file' at position 'sd->pos'
Jens Axboe016b6612006-04-25 15:42:00 +0200519 * using sendpage(). Return the number of bytes sent.
Jens Axboe5274f052006-03-30 15:15:30 +0200520 */
Jens Axboe76ad4d12006-05-03 10:41:33 +0200521static int pipe_to_sendpage(struct pipe_inode_info *pipe,
Jens Axboe5274f052006-03-30 15:15:30 +0200522 struct pipe_buffer *buf, struct splice_desc *sd)
523{
Jens Axboe6a14b902007-06-14 13:08:55 +0200524 struct file *file = sd->u.file;
Jens Axboe5274f052006-03-30 15:15:30 +0200525 loff_t pos = sd->pos;
Jens Axboef84d7512006-05-01 19:59:03 +0200526 int ret, more;
Jens Axboe5274f052006-03-30 15:15:30 +0200527
Jens Axboecac36bb2007-06-14 13:10:48 +0200528 ret = buf->ops->confirm(pipe, buf);
Jens Axboef84d7512006-05-01 19:59:03 +0200529 if (!ret) {
530 more = (sd->flags & SPLICE_F_MORE) || sd->len < sd->total_len;
Jens Axboe5274f052006-03-30 15:15:30 +0200531
Jens Axboef84d7512006-05-01 19:59:03 +0200532 ret = file->f_op->sendpage(file, buf->page, buf->offset,
533 sd->len, &pos, more);
534 }
Jens Axboe5274f052006-03-30 15:15:30 +0200535
Jens Axboe016b6612006-04-25 15:42:00 +0200536 return ret;
Jens Axboe5274f052006-03-30 15:15:30 +0200537}
538
539/*
540 * This is a little more tricky than the file -> pipe splicing. There are
541 * basically three cases:
542 *
543 * - Destination page already exists in the address space and there
544 * are users of it. For that case we have no other option that
545 * copying the data. Tough luck.
546 * - Destination page already exists in the address space, but there
547 * are no users of it. Make sure it's uptodate, then drop it. Fall
548 * through to last case.
549 * - Destination page does not exist, we can add the pipe page to
550 * the page cache and avoid the copy.
551 *
Jens Axboe83f91352006-04-02 23:05:09 +0200552 * If asked to move pages to the output file (SPLICE_F_MOVE is set in
553 * sd->flags), we attempt to migrate pages from the pipe to the output
554 * file address space page cache. This is possible if no one else has
555 * the pipe page referenced outside of the pipe and page cache. If
556 * SPLICE_F_MOVE isn't set, or we cannot move the page, we simply create
557 * a new page in the output file page cache and fill/dirty that.
Jens Axboe5274f052006-03-30 15:15:30 +0200558 */
Jens Axboe76ad4d12006-05-03 10:41:33 +0200559static int pipe_to_file(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
Jens Axboe5274f052006-03-30 15:15:30 +0200560 struct splice_desc *sd)
561{
Jens Axboe6a14b902007-06-14 13:08:55 +0200562 struct file *file = sd->u.file;
Jens Axboe5274f052006-03-30 15:15:30 +0200563 struct address_space *mapping = file->f_mapping;
Jens Axboe016b6612006-04-25 15:42:00 +0200564 unsigned int offset, this_len;
Jens Axboe5274f052006-03-30 15:15:30 +0200565 struct page *page;
Nick Pigginafddba42007-10-16 01:25:01 -0700566 void *fsdata;
Jens Axboe3e7ee3e2006-04-02 23:11:04 +0200567 int ret;
Jens Axboe5274f052006-03-30 15:15:30 +0200568
569 /*
Jens Axboe49d0b212006-04-10 09:04:41 +0200570 * make sure the data in this buffer is uptodate
Jens Axboe5274f052006-03-30 15:15:30 +0200571 */
Jens Axboecac36bb2007-06-14 13:10:48 +0200572 ret = buf->ops->confirm(pipe, buf);
Jens Axboef84d7512006-05-01 19:59:03 +0200573 if (unlikely(ret))
574 return ret;
Jens Axboe5274f052006-03-30 15:15:30 +0200575
Jens Axboe5274f052006-03-30 15:15:30 +0200576 offset = sd->pos & ~PAGE_CACHE_MASK;
577
Jens Axboe016b6612006-04-25 15:42:00 +0200578 this_len = sd->len;
579 if (this_len + offset > PAGE_CACHE_SIZE)
580 this_len = PAGE_CACHE_SIZE - offset;
581
Nick Pigginafddba42007-10-16 01:25:01 -0700582 ret = pagecache_write_begin(file, mapping, sd->pos, this_len,
583 AOP_FLAG_UNINTERRUPTIBLE, &page, &fsdata);
584 if (unlikely(ret))
585 goto out;
Jens Axboe5274f052006-03-30 15:15:30 +0200586
Jens Axboe0568b402006-05-01 19:50:48 +0200587 if (buf->page != page) {
Jens Axboef84d7512006-05-01 19:59:03 +0200588 /*
589 * Careful, ->map() uses KM_USER0!
590 */
Jens Axboe76ad4d12006-05-03 10:41:33 +0200591 char *src = buf->ops->map(pipe, buf, 1);
Jens Axboef84d7512006-05-01 19:59:03 +0200592 char *dst = kmap_atomic(page, KM_USER1);
Jens Axboe5abc97a2006-03-30 15:16:46 +0200593
Jens Axboe016b6612006-04-25 15:42:00 +0200594 memcpy(dst + offset, src + buf->offset, this_len);
Jens Axboe5abc97a2006-03-30 15:16:46 +0200595 flush_dcache_page(page);
Jens Axboef84d7512006-05-01 19:59:03 +0200596 kunmap_atomic(dst, KM_USER1);
Jens Axboe76ad4d12006-05-03 10:41:33 +0200597 buf->ops->unmap(pipe, buf, src);
Jens Axboe5abc97a2006-03-30 15:16:46 +0200598 }
Nick Pigginafddba42007-10-16 01:25:01 -0700599 ret = pagecache_write_end(file, mapping, sd->pos, this_len, this_len,
600 page, fsdata);
Jens Axboe5274f052006-03-30 15:15:30 +0200601out:
Jens Axboe5274f052006-03-30 15:15:30 +0200602 return ret;
603}
604
Jens Axboe932cc6d2007-06-21 13:10:21 +0200605/**
606 * __splice_from_pipe - splice data from a pipe to given actor
607 * @pipe: pipe to splice from
608 * @sd: information to @actor
609 * @actor: handler that splices the data
610 *
611 * Description:
612 * This function does little more than loop over the pipe and call
613 * @actor to do the actual moving of a single struct pipe_buffer to
614 * the desired destination. See pipe_to_file, pipe_to_sendpage, or
615 * pipe_to_user.
616 *
Jens Axboe83f91352006-04-02 23:05:09 +0200617 */
Jens Axboec66ab6f2007-06-12 21:17:17 +0200618ssize_t __splice_from_pipe(struct pipe_inode_info *pipe, struct splice_desc *sd,
619 splice_actor *actor)
Jens Axboe5274f052006-03-30 15:15:30 +0200620{
Jens Axboe5274f052006-03-30 15:15:30 +0200621 int ret, do_wakeup, err;
Jens Axboe5274f052006-03-30 15:15:30 +0200622
623 ret = 0;
624 do_wakeup = 0;
625
Jens Axboe5274f052006-03-30 15:15:30 +0200626 for (;;) {
Jens Axboe6f767b02006-04-11 13:53:56 +0200627 if (pipe->nrbufs) {
628 struct pipe_buffer *buf = pipe->bufs + pipe->curbuf;
Eric Dumazetd4c3cca2006-12-13 00:34:04 -0800629 const struct pipe_buf_operations *ops = buf->ops;
Jens Axboe5274f052006-03-30 15:15:30 +0200630
Jens Axboec66ab6f2007-06-12 21:17:17 +0200631 sd->len = buf->len;
632 if (sd->len > sd->total_len)
633 sd->len = sd->total_len;
Jens Axboe5274f052006-03-30 15:15:30 +0200634
Jens Axboec66ab6f2007-06-12 21:17:17 +0200635 err = actor(pipe, buf, sd);
Jens Axboe016b6612006-04-25 15:42:00 +0200636 if (err <= 0) {
Jens Axboe5274f052006-03-30 15:15:30 +0200637 if (!ret && err != -ENODATA)
638 ret = err;
639
640 break;
641 }
642
Jens Axboe016b6612006-04-25 15:42:00 +0200643 ret += err;
644 buf->offset += err;
645 buf->len -= err;
646
Jens Axboec66ab6f2007-06-12 21:17:17 +0200647 sd->len -= err;
648 sd->pos += err;
649 sd->total_len -= err;
650 if (sd->len)
Jens Axboe016b6612006-04-25 15:42:00 +0200651 continue;
Ingo Molnar73d62d82006-04-11 13:57:21 +0200652
Jens Axboe5274f052006-03-30 15:15:30 +0200653 if (!buf->len) {
654 buf->ops = NULL;
Ingo Molnar3a326a22006-04-10 15:18:35 +0200655 ops->release(pipe, buf);
Jens Axboe6f767b02006-04-11 13:53:56 +0200656 pipe->curbuf = (pipe->curbuf + 1) & (PIPE_BUFFERS - 1);
657 pipe->nrbufs--;
658 if (pipe->inode)
659 do_wakeup = 1;
Jens Axboe5274f052006-03-30 15:15:30 +0200660 }
661
Jens Axboec66ab6f2007-06-12 21:17:17 +0200662 if (!sd->total_len)
Jens Axboe5274f052006-03-30 15:15:30 +0200663 break;
664 }
665
Jens Axboe6f767b02006-04-11 13:53:56 +0200666 if (pipe->nrbufs)
Jens Axboe5274f052006-03-30 15:15:30 +0200667 continue;
Ingo Molnar3a326a22006-04-10 15:18:35 +0200668 if (!pipe->writers)
Jens Axboe5274f052006-03-30 15:15:30 +0200669 break;
Ingo Molnar3a326a22006-04-10 15:18:35 +0200670 if (!pipe->waiting_writers) {
Jens Axboe5274f052006-03-30 15:15:30 +0200671 if (ret)
672 break;
673 }
674
Jens Axboec66ab6f2007-06-12 21:17:17 +0200675 if (sd->flags & SPLICE_F_NONBLOCK) {
Linus Torvalds29e35092006-04-02 12:46:35 -0700676 if (!ret)
677 ret = -EAGAIN;
678 break;
679 }
680
Jens Axboe5274f052006-03-30 15:15:30 +0200681 if (signal_pending(current)) {
682 if (!ret)
683 ret = -ERESTARTSYS;
684 break;
685 }
686
687 if (do_wakeup) {
Jens Axboec0bd1f62006-04-10 09:03:32 +0200688 smp_mb();
Ingo Molnar3a326a22006-04-10 15:18:35 +0200689 if (waitqueue_active(&pipe->wait))
690 wake_up_interruptible_sync(&pipe->wait);
691 kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
Jens Axboe5274f052006-03-30 15:15:30 +0200692 do_wakeup = 0;
693 }
694
Ingo Molnar3a326a22006-04-10 15:18:35 +0200695 pipe_wait(pipe);
Jens Axboe5274f052006-03-30 15:15:30 +0200696 }
697
Jens Axboe5274f052006-03-30 15:15:30 +0200698 if (do_wakeup) {
Jens Axboec0bd1f62006-04-10 09:03:32 +0200699 smp_mb();
Ingo Molnar3a326a22006-04-10 15:18:35 +0200700 if (waitqueue_active(&pipe->wait))
701 wake_up_interruptible(&pipe->wait);
702 kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
Jens Axboe5274f052006-03-30 15:15:30 +0200703 }
704
Jens Axboe5274f052006-03-30 15:15:30 +0200705 return ret;
Jens Axboe5274f052006-03-30 15:15:30 +0200706}
Mark Fasheh40bee442007-03-21 13:11:02 +0100707EXPORT_SYMBOL(__splice_from_pipe);
Jens Axboe5274f052006-03-30 15:15:30 +0200708
Jens Axboe932cc6d2007-06-21 13:10:21 +0200709/**
710 * splice_from_pipe - splice data from a pipe to a file
711 * @pipe: pipe to splice from
712 * @out: file to splice to
713 * @ppos: position in @out
714 * @len: how many bytes to splice
715 * @flags: splice modifier flags
716 * @actor: handler that splices the data
717 *
718 * Description:
719 * See __splice_from_pipe. This function locks the input and output inodes,
720 * otherwise it's identical to __splice_from_pipe().
721 *
722 */
Mark Fasheh6da61802006-10-17 18:43:07 +0200723ssize_t splice_from_pipe(struct pipe_inode_info *pipe, struct file *out,
724 loff_t *ppos, size_t len, unsigned int flags,
725 splice_actor *actor)
726{
727 ssize_t ret;
728 struct inode *inode = out->f_mapping->host;
Jens Axboec66ab6f2007-06-12 21:17:17 +0200729 struct splice_desc sd = {
730 .total_len = len,
731 .flags = flags,
732 .pos = *ppos,
Jens Axboe6a14b902007-06-14 13:08:55 +0200733 .u.file = out,
Jens Axboec66ab6f2007-06-12 21:17:17 +0200734 };
Mark Fasheh6da61802006-10-17 18:43:07 +0200735
736 /*
737 * The actor worker might be calling ->prepare_write and
738 * ->commit_write. Most of the time, these expect i_mutex to
739 * be held. Since this may result in an ABBA deadlock with
740 * pipe->inode, we have to order lock acquiry here.
741 */
742 inode_double_lock(inode, pipe->inode);
Jens Axboec66ab6f2007-06-12 21:17:17 +0200743 ret = __splice_from_pipe(pipe, &sd, actor);
Mark Fasheh6da61802006-10-17 18:43:07 +0200744 inode_double_unlock(inode, pipe->inode);
745
746 return ret;
747}
748
749/**
750 * generic_file_splice_write_nolock - generic_file_splice_write without mutexes
751 * @pipe: pipe info
752 * @out: file to write to
Jens Axboe932cc6d2007-06-21 13:10:21 +0200753 * @ppos: position in @out
Mark Fasheh6da61802006-10-17 18:43:07 +0200754 * @len: number of bytes to splice
755 * @flags: splice modifier flags
756 *
Jens Axboe932cc6d2007-06-21 13:10:21 +0200757 * Description:
758 * Will either move or copy pages (determined by @flags options) from
759 * the given pipe inode to the given file. The caller is responsible
760 * for acquiring i_mutex on both inodes.
Mark Fasheh6da61802006-10-17 18:43:07 +0200761 *
762 */
763ssize_t
764generic_file_splice_write_nolock(struct pipe_inode_info *pipe, struct file *out,
765 loff_t *ppos, size_t len, unsigned int flags)
766{
767 struct address_space *mapping = out->f_mapping;
768 struct inode *inode = mapping->host;
Jens Axboec66ab6f2007-06-12 21:17:17 +0200769 struct splice_desc sd = {
770 .total_len = len,
771 .flags = flags,
772 .pos = *ppos,
Jens Axboe6a14b902007-06-14 13:08:55 +0200773 .u.file = out,
Jens Axboec66ab6f2007-06-12 21:17:17 +0200774 };
Mark Fasheh6da61802006-10-17 18:43:07 +0200775 ssize_t ret;
776 int err;
777
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800778 err = remove_suid(out->f_path.dentry);
Jens Axboe8c34e2d2006-10-17 19:43:22 +0200779 if (unlikely(err))
780 return err;
781
Jens Axboec66ab6f2007-06-12 21:17:17 +0200782 ret = __splice_from_pipe(pipe, &sd, pipe_to_file);
Mark Fasheh6da61802006-10-17 18:43:07 +0200783 if (ret > 0) {
Jens Axboe17ee4f42007-06-15 13:10:37 +0200784 unsigned long nr_pages;
785
Mark Fasheh6da61802006-10-17 18:43:07 +0200786 *ppos += ret;
Jens Axboe17ee4f42007-06-15 13:10:37 +0200787 nr_pages = (ret + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
Mark Fasheh6da61802006-10-17 18:43:07 +0200788
789 /*
790 * If file or inode is SYNC and we actually wrote some data,
791 * sync it.
792 */
793 if (unlikely((out->f_flags & O_SYNC) || IS_SYNC(inode))) {
794 err = generic_osync_inode(inode, mapping,
795 OSYNC_METADATA|OSYNC_DATA);
796
797 if (err)
798 ret = err;
799 }
Jens Axboe17ee4f42007-06-15 13:10:37 +0200800 balance_dirty_pages_ratelimited_nr(mapping, nr_pages);
Mark Fasheh6da61802006-10-17 18:43:07 +0200801 }
802
803 return ret;
804}
805
806EXPORT_SYMBOL(generic_file_splice_write_nolock);
807
Jens Axboe83f91352006-04-02 23:05:09 +0200808/**
809 * generic_file_splice_write - splice data from a pipe to a file
Ingo Molnar3a326a22006-04-10 15:18:35 +0200810 * @pipe: pipe info
Jens Axboe83f91352006-04-02 23:05:09 +0200811 * @out: file to write to
Jens Axboe932cc6d2007-06-21 13:10:21 +0200812 * @ppos: position in @out
Jens Axboe83f91352006-04-02 23:05:09 +0200813 * @len: number of bytes to splice
814 * @flags: splice modifier flags
815 *
Jens Axboe932cc6d2007-06-21 13:10:21 +0200816 * Description:
817 * Will either move or copy pages (determined by @flags options) from
818 * the given pipe inode to the given file.
Jens Axboe83f91352006-04-02 23:05:09 +0200819 *
820 */
Ingo Molnar3a326a22006-04-10 15:18:35 +0200821ssize_t
822generic_file_splice_write(struct pipe_inode_info *pipe, struct file *out,
Jens Axboecbb7e572006-04-11 14:57:50 +0200823 loff_t *ppos, size_t len, unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +0200824{
Jens Axboe4f6f0bd2006-04-02 23:04:46 +0200825 struct address_space *mapping = out->f_mapping;
Jens Axboe8c34e2d2006-10-17 19:43:22 +0200826 struct inode *inode = mapping->host;
Ingo Molnar3a326a22006-04-10 15:18:35 +0200827 ssize_t ret;
Jens Axboe8c34e2d2006-10-17 19:43:22 +0200828 int err;
829
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800830 err = should_remove_suid(out->f_path.dentry);
Jens Axboe8c34e2d2006-10-17 19:43:22 +0200831 if (unlikely(err)) {
832 mutex_lock(&inode->i_mutex);
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800833 err = __remove_suid(out->f_path.dentry, err);
Jens Axboe8c34e2d2006-10-17 19:43:22 +0200834 mutex_unlock(&inode->i_mutex);
835 if (err)
836 return err;
837 }
Ingo Molnar3a326a22006-04-10 15:18:35 +0200838
Jens Axboe00522fb2006-04-26 14:39:29 +0200839 ret = splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_file);
Jens Axboea4514eb2006-04-19 15:57:05 +0200840 if (ret > 0) {
Jens Axboe17ee4f42007-06-15 13:10:37 +0200841 unsigned long nr_pages;
842
Jens Axboea4514eb2006-04-19 15:57:05 +0200843 *ppos += ret;
Jens Axboe17ee4f42007-06-15 13:10:37 +0200844 nr_pages = (ret + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
Jens Axboe4f6f0bd2006-04-02 23:04:46 +0200845
Jens Axboea4514eb2006-04-19 15:57:05 +0200846 /*
847 * If file or inode is SYNC and we actually wrote some data,
848 * sync it.
849 */
850 if (unlikely((out->f_flags & O_SYNC) || IS_SYNC(inode))) {
Jens Axboea4514eb2006-04-19 15:57:05 +0200851 mutex_lock(&inode->i_mutex);
852 err = generic_osync_inode(inode, mapping,
853 OSYNC_METADATA|OSYNC_DATA);
854 mutex_unlock(&inode->i_mutex);
855
856 if (err)
857 ret = err;
858 }
Jens Axboe17ee4f42007-06-15 13:10:37 +0200859 balance_dirty_pages_ratelimited_nr(mapping, nr_pages);
Jens Axboe4f6f0bd2006-04-02 23:04:46 +0200860 }
861
862 return ret;
Jens Axboe5274f052006-03-30 15:15:30 +0200863}
864
Jens Axboe059a8f32006-04-02 23:06:05 +0200865EXPORT_SYMBOL(generic_file_splice_write);
866
Jens Axboe83f91352006-04-02 23:05:09 +0200867/**
868 * generic_splice_sendpage - splice data from a pipe to a socket
Jens Axboe932cc6d2007-06-21 13:10:21 +0200869 * @pipe: pipe to splice from
Jens Axboe83f91352006-04-02 23:05:09 +0200870 * @out: socket to write to
Jens Axboe932cc6d2007-06-21 13:10:21 +0200871 * @ppos: position in @out
Jens Axboe83f91352006-04-02 23:05:09 +0200872 * @len: number of bytes to splice
873 * @flags: splice modifier flags
874 *
Jens Axboe932cc6d2007-06-21 13:10:21 +0200875 * Description:
876 * Will send @len bytes from the pipe to a network socket. No data copying
877 * is involved.
Jens Axboe83f91352006-04-02 23:05:09 +0200878 *
879 */
Ingo Molnar3a326a22006-04-10 15:18:35 +0200880ssize_t generic_splice_sendpage(struct pipe_inode_info *pipe, struct file *out,
Jens Axboecbb7e572006-04-11 14:57:50 +0200881 loff_t *ppos, size_t len, unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +0200882{
Jens Axboe00522fb2006-04-26 14:39:29 +0200883 return splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_sendpage);
Jens Axboe5274f052006-03-30 15:15:30 +0200884}
885
Jens Axboe059a8f32006-04-02 23:06:05 +0200886EXPORT_SYMBOL(generic_splice_sendpage);
Jeff Garzika0f06782006-03-30 23:06:13 -0500887
Jens Axboe83f91352006-04-02 23:05:09 +0200888/*
889 * Attempt to initiate a splice from pipe to file.
890 */
Ingo Molnar3a326a22006-04-10 15:18:35 +0200891static long do_splice_from(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 Axboe5274f052006-03-30 15:15:30 +0200894 int ret;
895
Jens Axboe49570e92006-04-11 13:56:09 +0200896 if (unlikely(!out->f_op || !out->f_op->splice_write))
Jens Axboe5274f052006-03-30 15:15:30 +0200897 return -EINVAL;
898
Jens Axboe49570e92006-04-11 13:56:09 +0200899 if (unlikely(!(out->f_mode & FMODE_WRITE)))
Jens Axboe5274f052006-03-30 15:15:30 +0200900 return -EBADF;
901
Jens Axboecbb7e572006-04-11 14:57:50 +0200902 ret = rw_verify_area(WRITE, out, ppos, len);
Jens Axboe5274f052006-03-30 15:15:30 +0200903 if (unlikely(ret < 0))
904 return ret;
905
James Morris29ce2052007-07-13 11:44:32 +0200906 ret = security_file_permission(out, MAY_WRITE);
907 if (unlikely(ret < 0))
908 return ret;
909
Jens Axboecbb7e572006-04-11 14:57:50 +0200910 return out->f_op->splice_write(pipe, out, ppos, len, flags);
Jens Axboe5274f052006-03-30 15:15:30 +0200911}
912
Jens Axboe83f91352006-04-02 23:05:09 +0200913/*
914 * Attempt to initiate a splice from a file to a pipe.
915 */
Jens Axboecbb7e572006-04-11 14:57:50 +0200916static long do_splice_to(struct file *in, loff_t *ppos,
917 struct pipe_inode_info *pipe, size_t len,
918 unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +0200919{
Jens Axboe5274f052006-03-30 15:15:30 +0200920 int ret;
921
Jens Axboe49570e92006-04-11 13:56:09 +0200922 if (unlikely(!in->f_op || !in->f_op->splice_read))
Jens Axboe5274f052006-03-30 15:15:30 +0200923 return -EINVAL;
924
Jens Axboe49570e92006-04-11 13:56:09 +0200925 if (unlikely(!(in->f_mode & FMODE_READ)))
Jens Axboe5274f052006-03-30 15:15:30 +0200926 return -EBADF;
927
Jens Axboecbb7e572006-04-11 14:57:50 +0200928 ret = rw_verify_area(READ, in, ppos, len);
Jens Axboe5274f052006-03-30 15:15:30 +0200929 if (unlikely(ret < 0))
930 return ret;
931
James Morris29ce2052007-07-13 11:44:32 +0200932 ret = security_file_permission(in, MAY_READ);
933 if (unlikely(ret < 0))
934 return ret;
935
Jens Axboecbb7e572006-04-11 14:57:50 +0200936 return in->f_op->splice_read(in, ppos, pipe, len, flags);
Jens Axboe5274f052006-03-30 15:15:30 +0200937}
938
Jens Axboe932cc6d2007-06-21 13:10:21 +0200939/**
940 * splice_direct_to_actor - splices data directly between two non-pipes
941 * @in: file to splice from
942 * @sd: actor information on where to splice to
943 * @actor: handles the data splicing
944 *
945 * Description:
946 * This is a special case helper to splice directly between two
947 * points, without requiring an explicit pipe. Internally an allocated
Randy Dunlap79685b82007-07-27 08:08:51 +0200948 * pipe is cached in the process, and reused during the lifetime of
Jens Axboe932cc6d2007-06-21 13:10:21 +0200949 * that process.
950 *
Jens Axboec66ab6f2007-06-12 21:17:17 +0200951 */
952ssize_t splice_direct_to_actor(struct file *in, struct splice_desc *sd,
953 splice_direct_actor *actor)
Jens Axboeb92ce552006-04-11 13:52:07 +0200954{
955 struct pipe_inode_info *pipe;
956 long ret, bytes;
957 umode_t i_mode;
Jens Axboec66ab6f2007-06-12 21:17:17 +0200958 size_t len;
959 int i, flags;
Jens Axboeb92ce552006-04-11 13:52:07 +0200960
961 /*
962 * We require the input being a regular file, as we don't want to
963 * randomly drop data for eg socket -> socket splicing. Use the
964 * piped splicing for that!
965 */
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800966 i_mode = in->f_path.dentry->d_inode->i_mode;
Jens Axboeb92ce552006-04-11 13:52:07 +0200967 if (unlikely(!S_ISREG(i_mode) && !S_ISBLK(i_mode)))
968 return -EINVAL;
969
970 /*
971 * neither in nor out is a pipe, setup an internal pipe attached to
972 * 'out' and transfer the wanted data from 'in' to 'out' through that
973 */
974 pipe = current->splice_pipe;
Jens Axboe49570e92006-04-11 13:56:09 +0200975 if (unlikely(!pipe)) {
Jens Axboeb92ce552006-04-11 13:52:07 +0200976 pipe = alloc_pipe_info(NULL);
977 if (!pipe)
978 return -ENOMEM;
979
980 /*
981 * We don't have an immediate reader, but we'll read the stuff
Jens Axboe00522fb2006-04-26 14:39:29 +0200982 * out of the pipe right after the splice_to_pipe(). So set
Jens Axboeb92ce552006-04-11 13:52:07 +0200983 * PIPE_READERS appropriately.
984 */
985 pipe->readers = 1;
986
987 current->splice_pipe = pipe;
988 }
989
990 /*
Ingo Molnar73d62d82006-04-11 13:57:21 +0200991 * Do the splice.
Jens Axboeb92ce552006-04-11 13:52:07 +0200992 */
993 ret = 0;
994 bytes = 0;
Jens Axboec66ab6f2007-06-12 21:17:17 +0200995 len = sd->total_len;
996 flags = sd->flags;
997
998 /*
999 * Don't block on output, we have to drain the direct pipe.
1000 */
1001 sd->flags &= ~SPLICE_F_NONBLOCK;
Jens Axboeb92ce552006-04-11 13:52:07 +02001002
1003 while (len) {
Jens Axboe51a92c02007-07-13 14:11:43 +02001004 size_t read_len;
Jens Axboebcd4f3a2007-07-16 14:41:49 +02001005 loff_t pos = sd->pos;
Jens Axboeb92ce552006-04-11 13:52:07 +02001006
Jens Axboebcd4f3a2007-07-16 14:41:49 +02001007 ret = do_splice_to(in, &pos, pipe, len, flags);
Jens Axboe51a92c02007-07-13 14:11:43 +02001008 if (unlikely(ret <= 0))
Jens Axboeb92ce552006-04-11 13:52:07 +02001009 goto out_release;
1010
1011 read_len = ret;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001012 sd->total_len = read_len;
Jens Axboeb92ce552006-04-11 13:52:07 +02001013
1014 /*
1015 * NOTE: nonblocking mode only applies to the input. We
1016 * must not do the output in nonblocking mode as then we
1017 * could get stuck data in the internal pipe:
1018 */
Jens Axboec66ab6f2007-06-12 21:17:17 +02001019 ret = actor(pipe, sd);
Jens Axboe51a92c02007-07-13 14:11:43 +02001020 if (unlikely(ret <= 0))
Jens Axboeb92ce552006-04-11 13:52:07 +02001021 goto out_release;
1022
1023 bytes += ret;
1024 len -= ret;
Jens Axboebcd4f3a2007-07-16 14:41:49 +02001025 sd->pos = pos;
Jens Axboeb92ce552006-04-11 13:52:07 +02001026
Jens Axboe51a92c02007-07-13 14:11:43 +02001027 if (ret < read_len)
1028 goto out_release;
Jens Axboeb92ce552006-04-11 13:52:07 +02001029 }
1030
1031 pipe->nrbufs = pipe->curbuf = 0;
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 }
1047 pipe->nrbufs = pipe->curbuf = 0;
1048
1049 /*
1050 * If we transferred some data, return the number of bytes:
1051 */
1052 if (bytes > 0)
1053 return bytes;
1054
1055 return ret;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001056
1057}
1058EXPORT_SYMBOL(splice_direct_to_actor);
1059
1060static int direct_splice_actor(struct pipe_inode_info *pipe,
1061 struct splice_desc *sd)
1062{
Jens Axboe6a14b902007-06-14 13:08:55 +02001063 struct file *file = sd->u.file;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001064
1065 return do_splice_from(pipe, file, &sd->pos, sd->total_len, sd->flags);
1066}
1067
Jens Axboe932cc6d2007-06-21 13:10:21 +02001068/**
1069 * do_splice_direct - splices data directly between two files
1070 * @in: file to splice from
1071 * @ppos: input file offset
1072 * @out: file to splice to
1073 * @len: number of bytes to splice
1074 * @flags: splice modifier flags
1075 *
1076 * Description:
1077 * For use by do_sendfile(). splice can easily emulate sendfile, but
1078 * doing it in the application would incur an extra system call
1079 * (splice in + splice out, as compared to just sendfile()). So this helper
1080 * can splice directly through a process-private pipe.
1081 *
1082 */
Jens Axboec66ab6f2007-06-12 21:17:17 +02001083long do_splice_direct(struct file *in, loff_t *ppos, struct file *out,
1084 size_t len, unsigned int flags)
1085{
1086 struct splice_desc sd = {
1087 .len = len,
1088 .total_len = len,
1089 .flags = flags,
1090 .pos = *ppos,
Jens Axboe6a14b902007-06-14 13:08:55 +02001091 .u.file = out,
Jens Axboec66ab6f2007-06-12 21:17:17 +02001092 };
Jens Axboe51a92c02007-07-13 14:11:43 +02001093 long ret;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001094
1095 ret = splice_direct_to_actor(in, &sd, direct_splice_actor);
Jens Axboe51a92c02007-07-13 14:11:43 +02001096 if (ret > 0)
1097 *ppos += ret;
1098
Jens Axboec66ab6f2007-06-12 21:17:17 +02001099 return ret;
Jens Axboeb92ce552006-04-11 13:52:07 +02001100}
1101
Jens Axboe83f91352006-04-02 23:05:09 +02001102/*
Jens Axboeddac0d32006-11-04 12:49:32 +01001103 * After the inode slimming patch, i_pipe/i_bdev/i_cdev share the same
1104 * location, so checking ->i_pipe is not enough to verify that this is a
1105 * pipe.
1106 */
1107static inline struct pipe_inode_info *pipe_info(struct inode *inode)
1108{
1109 if (S_ISFIFO(inode->i_mode))
1110 return inode->i_pipe;
1111
1112 return NULL;
1113}
1114
1115/*
Jens Axboe83f91352006-04-02 23:05:09 +02001116 * Determine where to splice to/from.
1117 */
Ingo Molnar529565d2006-04-10 15:18:58 +02001118static long do_splice(struct file *in, loff_t __user *off_in,
1119 struct file *out, loff_t __user *off_out,
1120 size_t len, unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +02001121{
Ingo Molnar3a326a22006-04-10 15:18:35 +02001122 struct pipe_inode_info *pipe;
Jens Axboecbb7e572006-04-11 14:57:50 +02001123 loff_t offset, *off;
Jens Axboea4514eb2006-04-19 15:57:05 +02001124 long ret;
Jens Axboe5274f052006-03-30 15:15:30 +02001125
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001126 pipe = pipe_info(in->f_path.dentry->d_inode);
Ingo Molnar529565d2006-04-10 15:18:58 +02001127 if (pipe) {
1128 if (off_in)
1129 return -ESPIPE;
Jens Axboeb92ce552006-04-11 13:52:07 +02001130 if (off_out) {
1131 if (out->f_op->llseek == no_llseek)
1132 return -EINVAL;
Jens Axboecbb7e572006-04-11 14:57:50 +02001133 if (copy_from_user(&offset, off_out, sizeof(loff_t)))
Jens Axboeb92ce552006-04-11 13:52:07 +02001134 return -EFAULT;
Jens Axboecbb7e572006-04-11 14:57:50 +02001135 off = &offset;
1136 } else
1137 off = &out->f_pos;
Ingo Molnar529565d2006-04-10 15:18:58 +02001138
Jens Axboea4514eb2006-04-19 15:57:05 +02001139 ret = do_splice_from(pipe, out, off, len, flags);
1140
1141 if (off_out && copy_to_user(off_out, off, sizeof(loff_t)))
1142 ret = -EFAULT;
1143
1144 return ret;
Ingo Molnar529565d2006-04-10 15:18:58 +02001145 }
Jens Axboe5274f052006-03-30 15:15:30 +02001146
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001147 pipe = pipe_info(out->f_path.dentry->d_inode);
Ingo Molnar529565d2006-04-10 15:18:58 +02001148 if (pipe) {
1149 if (off_out)
1150 return -ESPIPE;
Jens Axboeb92ce552006-04-11 13:52:07 +02001151 if (off_in) {
1152 if (in->f_op->llseek == no_llseek)
1153 return -EINVAL;
Jens Axboecbb7e572006-04-11 14:57:50 +02001154 if (copy_from_user(&offset, off_in, sizeof(loff_t)))
Jens Axboeb92ce552006-04-11 13:52:07 +02001155 return -EFAULT;
Jens Axboecbb7e572006-04-11 14:57:50 +02001156 off = &offset;
1157 } else
1158 off = &in->f_pos;
Ingo Molnar529565d2006-04-10 15:18:58 +02001159
Jens Axboea4514eb2006-04-19 15:57:05 +02001160 ret = do_splice_to(in, off, pipe, len, flags);
1161
1162 if (off_in && copy_to_user(off_in, off, sizeof(loff_t)))
1163 ret = -EFAULT;
1164
1165 return ret;
Ingo Molnar529565d2006-04-10 15:18:58 +02001166 }
Jens Axboe5274f052006-03-30 15:15:30 +02001167
1168 return -EINVAL;
1169}
1170
Jens Axboe912d35f2006-04-26 10:59:21 +02001171/*
Linus Torvalds75723952007-10-01 13:17:28 -07001172 * Do a copy-from-user while holding the mmap_semaphore for reading, in a
1173 * manner safe from deadlocking with simultaneous mmap() (grabbing mmap_sem
1174 * for writing) and page faulting on the user memory pointed to by src.
1175 * This assumes that we will very rarely hit the partial != 0 path, or this
1176 * will not be a win.
1177 */
1178static int copy_from_user_mmap_sem(void *dst, const void __user *src, size_t n)
1179{
1180 int partial;
1181
1182 pagefault_disable();
1183 partial = __copy_from_user_inatomic(dst, src, n);
1184 pagefault_enable();
1185
1186 /*
1187 * Didn't copy everything, drop the mmap_sem and do a faulting copy
1188 */
1189 if (unlikely(partial)) {
1190 up_read(&current->mm->mmap_sem);
1191 partial = copy_from_user(dst, src, n);
1192 down_read(&current->mm->mmap_sem);
1193 }
1194
1195 return partial;
1196}
1197
1198/*
Jens Axboe912d35f2006-04-26 10:59:21 +02001199 * Map an iov into an array of pages and offset/length tupples. With the
1200 * partial_page structure, we can map several non-contiguous ranges into
1201 * our ones pages[] map instead of splitting that operation into pieces.
1202 * Could easily be exported as a generic helper for other users, in which
1203 * case one would probably want to add a 'max_nr_pages' parameter as well.
1204 */
1205static int get_iovec_page_array(const struct iovec __user *iov,
1206 unsigned int nr_vecs, struct page **pages,
Jens Axboe7afa6fd2006-05-01 20:02:33 +02001207 struct partial_page *partial, int aligned)
Jens Axboe912d35f2006-04-26 10:59:21 +02001208{
1209 int buffers = 0, error = 0;
1210
Jens Axboe912d35f2006-04-26 10:59:21 +02001211 down_read(&current->mm->mmap_sem);
1212
1213 while (nr_vecs) {
1214 unsigned long off, npages;
Linus Torvalds75723952007-10-01 13:17:28 -07001215 struct iovec entry;
Jens Axboe912d35f2006-04-26 10:59:21 +02001216 void __user *base;
1217 size_t len;
1218 int i;
1219
Linus Torvalds75723952007-10-01 13:17:28 -07001220 error = -EFAULT;
1221 if (copy_from_user_mmap_sem(&entry, iov, sizeof(entry)))
Jens Axboe912d35f2006-04-26 10:59:21 +02001222 break;
Linus Torvalds75723952007-10-01 13:17:28 -07001223
1224 base = entry.iov_base;
1225 len = entry.iov_len;
Jens Axboe912d35f2006-04-26 10:59:21 +02001226
1227 /*
1228 * Sanity check this iovec. 0 read succeeds.
1229 */
Linus Torvalds75723952007-10-01 13:17:28 -07001230 error = 0;
Jens Axboe912d35f2006-04-26 10:59:21 +02001231 if (unlikely(!len))
1232 break;
1233 error = -EFAULT;
1234 if (unlikely(!base))
1235 break;
1236
1237 /*
1238 * Get this base offset and number of pages, then map
1239 * in the user pages.
1240 */
1241 off = (unsigned long) base & ~PAGE_MASK;
Jens Axboe7afa6fd2006-05-01 20:02:33 +02001242
1243 /*
1244 * If asked for alignment, the offset must be zero and the
1245 * length a multiple of the PAGE_SIZE.
1246 */
1247 error = -EINVAL;
1248 if (aligned && (off || len & ~PAGE_MASK))
1249 break;
1250
Jens Axboe912d35f2006-04-26 10:59:21 +02001251 npages = (off + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1252 if (npages > PIPE_BUFFERS - buffers)
1253 npages = PIPE_BUFFERS - buffers;
1254
1255 error = get_user_pages(current, current->mm,
1256 (unsigned long) base, npages, 0, 0,
1257 &pages[buffers], NULL);
1258
1259 if (unlikely(error <= 0))
1260 break;
1261
1262 /*
1263 * Fill this contiguous range into the partial page map.
1264 */
1265 for (i = 0; i < error; i++) {
Jens Axboe75914892006-05-02 12:57:18 +02001266 const int plen = min_t(size_t, len, PAGE_SIZE - off);
Jens Axboe912d35f2006-04-26 10:59:21 +02001267
1268 partial[buffers].offset = off;
1269 partial[buffers].len = plen;
1270
1271 off = 0;
1272 len -= plen;
1273 buffers++;
1274 }
1275
1276 /*
1277 * We didn't complete this iov, stop here since it probably
1278 * means we have to move some of this into a pipe to
1279 * be able to continue.
1280 */
1281 if (len)
1282 break;
1283
1284 /*
1285 * Don't continue if we mapped fewer pages than we asked for,
1286 * or if we mapped the max number of pages that we have
1287 * room for.
1288 */
1289 if (error < npages || buffers == PIPE_BUFFERS)
1290 break;
1291
1292 nr_vecs--;
1293 iov++;
1294 }
1295
1296 up_read(&current->mm->mmap_sem);
1297
1298 if (buffers)
1299 return buffers;
1300
1301 return error;
1302}
1303
Jens Axboe6a14b902007-06-14 13:08:55 +02001304static int pipe_to_user(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
1305 struct splice_desc *sd)
1306{
1307 char *src;
1308 int ret;
1309
Jens Axboecac36bb2007-06-14 13:10:48 +02001310 ret = buf->ops->confirm(pipe, buf);
Jens Axboe6a14b902007-06-14 13:08:55 +02001311 if (unlikely(ret))
1312 return ret;
1313
1314 /*
1315 * See if we can use the atomic maps, by prefaulting in the
1316 * pages and doing an atomic copy
1317 */
1318 if (!fault_in_pages_writeable(sd->u.userptr, sd->len)) {
1319 src = buf->ops->map(pipe, buf, 1);
1320 ret = __copy_to_user_inatomic(sd->u.userptr, src + buf->offset,
1321 sd->len);
1322 buf->ops->unmap(pipe, buf, src);
1323 if (!ret) {
1324 ret = sd->len;
1325 goto out;
1326 }
1327 }
1328
1329 /*
1330 * No dice, use slow non-atomic map and copy
1331 */
1332 src = buf->ops->map(pipe, buf, 0);
1333
1334 ret = sd->len;
1335 if (copy_to_user(sd->u.userptr, src + buf->offset, sd->len))
1336 ret = -EFAULT;
1337
Jens Axboe6866bef2007-10-16 10:01:29 +02001338 buf->ops->unmap(pipe, buf, src);
Jens Axboe6a14b902007-06-14 13:08:55 +02001339out:
1340 if (ret > 0)
1341 sd->u.userptr += ret;
Jens Axboe6a14b902007-06-14 13:08:55 +02001342 return ret;
1343}
1344
1345/*
1346 * For lack of a better implementation, implement vmsplice() to userspace
1347 * as a simple copy of the pipes pages to the user iov.
1348 */
1349static long vmsplice_to_user(struct file *file, const struct iovec __user *iov,
1350 unsigned long nr_segs, unsigned int flags)
1351{
1352 struct pipe_inode_info *pipe;
1353 struct splice_desc sd;
1354 ssize_t size;
1355 int error;
1356 long ret;
1357
1358 pipe = pipe_info(file->f_path.dentry->d_inode);
1359 if (!pipe)
1360 return -EBADF;
1361
1362 if (pipe->inode)
1363 mutex_lock(&pipe->inode->i_mutex);
1364
1365 error = ret = 0;
1366 while (nr_segs) {
1367 void __user *base;
1368 size_t len;
1369
1370 /*
1371 * Get user address base and length for this iovec.
1372 */
1373 error = get_user(base, &iov->iov_base);
1374 if (unlikely(error))
1375 break;
1376 error = get_user(len, &iov->iov_len);
1377 if (unlikely(error))
1378 break;
1379
1380 /*
1381 * Sanity check this iovec. 0 read succeeds.
1382 */
1383 if (unlikely(!len))
1384 break;
1385 if (unlikely(!base)) {
1386 error = -EFAULT;
1387 break;
1388 }
1389
1390 sd.len = 0;
1391 sd.total_len = len;
1392 sd.flags = flags;
1393 sd.u.userptr = base;
1394 sd.pos = 0;
1395
1396 size = __splice_from_pipe(pipe, &sd, pipe_to_user);
1397 if (size < 0) {
1398 if (!ret)
1399 ret = size;
1400
1401 break;
1402 }
1403
1404 ret += size;
1405
1406 if (size < len)
1407 break;
1408
1409 nr_segs--;
1410 iov++;
1411 }
1412
1413 if (pipe->inode)
1414 mutex_unlock(&pipe->inode->i_mutex);
1415
1416 if (!ret)
1417 ret = error;
1418
1419 return ret;
1420}
1421
Jens Axboe912d35f2006-04-26 10:59:21 +02001422/*
1423 * vmsplice splices a user address range into a pipe. It can be thought of
1424 * as splice-from-memory, where the regular splice is splice-from-file (or
1425 * to file). In both cases the output is a pipe, naturally.
Jens Axboe912d35f2006-04-26 10:59:21 +02001426 */
Jens Axboe6a14b902007-06-14 13:08:55 +02001427static long vmsplice_to_pipe(struct file *file, const struct iovec __user *iov,
1428 unsigned long nr_segs, unsigned int flags)
Jens Axboe912d35f2006-04-26 10:59:21 +02001429{
Jens Axboeddac0d32006-11-04 12:49:32 +01001430 struct pipe_inode_info *pipe;
Jens Axboe912d35f2006-04-26 10:59:21 +02001431 struct page *pages[PIPE_BUFFERS];
1432 struct partial_page partial[PIPE_BUFFERS];
1433 struct splice_pipe_desc spd = {
1434 .pages = pages,
1435 .partial = partial,
1436 .flags = flags,
1437 .ops = &user_page_pipe_buf_ops,
1438 };
1439
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001440 pipe = pipe_info(file->f_path.dentry->d_inode);
Jens Axboeddac0d32006-11-04 12:49:32 +01001441 if (!pipe)
Jens Axboe912d35f2006-04-26 10:59:21 +02001442 return -EBADF;
Jens Axboe912d35f2006-04-26 10:59:21 +02001443
Jens Axboe7afa6fd2006-05-01 20:02:33 +02001444 spd.nr_pages = get_iovec_page_array(iov, nr_segs, pages, partial,
1445 flags & SPLICE_F_GIFT);
Jens Axboe912d35f2006-04-26 10:59:21 +02001446 if (spd.nr_pages <= 0)
1447 return spd.nr_pages;
1448
Jens Axboe00522fb2006-04-26 14:39:29 +02001449 return splice_to_pipe(pipe, &spd);
Jens Axboe912d35f2006-04-26 10:59:21 +02001450}
1451
Jens Axboe6a14b902007-06-14 13:08:55 +02001452/*
1453 * Note that vmsplice only really supports true splicing _from_ user memory
1454 * to a pipe, not the other way around. Splicing from user memory is a simple
1455 * operation that can be supported without any funky alignment restrictions
1456 * or nasty vm tricks. We simply map in the user memory and fill them into
1457 * a pipe. The reverse isn't quite as easy, though. There are two possible
1458 * solutions for that:
1459 *
1460 * - memcpy() the data internally, at which point we might as well just
1461 * do a regular read() on the buffer anyway.
1462 * - Lots of nasty vm tricks, that are neither fast nor flexible (it
1463 * has restriction limitations on both ends of the pipe).
1464 *
1465 * Currently we punt and implement it as a normal copy, see pipe_to_user().
1466 *
1467 */
Jens Axboe912d35f2006-04-26 10:59:21 +02001468asmlinkage long sys_vmsplice(int fd, const struct iovec __user *iov,
1469 unsigned long nr_segs, unsigned int flags)
1470{
1471 struct file *file;
1472 long error;
1473 int fput;
1474
Jens Axboe6a14b902007-06-14 13:08:55 +02001475 if (unlikely(nr_segs > UIO_MAXIOV))
1476 return -EINVAL;
1477 else if (unlikely(!nr_segs))
1478 return 0;
1479
Jens Axboe912d35f2006-04-26 10:59:21 +02001480 error = -EBADF;
1481 file = fget_light(fd, &fput);
1482 if (file) {
1483 if (file->f_mode & FMODE_WRITE)
Jens Axboe6a14b902007-06-14 13:08:55 +02001484 error = vmsplice_to_pipe(file, iov, nr_segs, flags);
1485 else if (file->f_mode & FMODE_READ)
1486 error = vmsplice_to_user(file, iov, nr_segs, flags);
Jens Axboe912d35f2006-04-26 10:59:21 +02001487
1488 fput_light(file, fput);
1489 }
1490
1491 return error;
1492}
1493
Ingo Molnar529565d2006-04-10 15:18:58 +02001494asmlinkage long sys_splice(int fd_in, loff_t __user *off_in,
1495 int fd_out, loff_t __user *off_out,
1496 size_t len, unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +02001497{
1498 long error;
1499 struct file *in, *out;
1500 int fput_in, fput_out;
1501
1502 if (unlikely(!len))
1503 return 0;
1504
1505 error = -EBADF;
Ingo Molnar529565d2006-04-10 15:18:58 +02001506 in = fget_light(fd_in, &fput_in);
Jens Axboe5274f052006-03-30 15:15:30 +02001507 if (in) {
1508 if (in->f_mode & FMODE_READ) {
Ingo Molnar529565d2006-04-10 15:18:58 +02001509 out = fget_light(fd_out, &fput_out);
Jens Axboe5274f052006-03-30 15:15:30 +02001510 if (out) {
1511 if (out->f_mode & FMODE_WRITE)
Ingo Molnar529565d2006-04-10 15:18:58 +02001512 error = do_splice(in, off_in,
1513 out, off_out,
1514 len, flags);
Jens Axboe5274f052006-03-30 15:15:30 +02001515 fput_light(out, fput_out);
1516 }
1517 }
1518
1519 fput_light(in, fput_in);
1520 }
1521
1522 return error;
1523}
Jens Axboe70524492006-04-11 15:51:17 +02001524
1525/*
Jens Axboeaadd06e2006-07-10 11:00:01 +02001526 * Make sure there's data to read. Wait for input if we can, otherwise
1527 * return an appropriate error.
1528 */
1529static int link_ipipe_prep(struct pipe_inode_info *pipe, unsigned int flags)
1530{
1531 int ret;
1532
1533 /*
1534 * Check ->nrbufs without the inode lock first. This function
1535 * is speculative anyways, so missing one is ok.
1536 */
1537 if (pipe->nrbufs)
1538 return 0;
1539
1540 ret = 0;
1541 mutex_lock(&pipe->inode->i_mutex);
1542
1543 while (!pipe->nrbufs) {
1544 if (signal_pending(current)) {
1545 ret = -ERESTARTSYS;
1546 break;
1547 }
1548 if (!pipe->writers)
1549 break;
1550 if (!pipe->waiting_writers) {
1551 if (flags & SPLICE_F_NONBLOCK) {
1552 ret = -EAGAIN;
1553 break;
1554 }
1555 }
1556 pipe_wait(pipe);
1557 }
1558
1559 mutex_unlock(&pipe->inode->i_mutex);
1560 return ret;
1561}
1562
1563/*
1564 * Make sure there's writeable room. Wait for room if we can, otherwise
1565 * return an appropriate error.
1566 */
1567static int link_opipe_prep(struct pipe_inode_info *pipe, unsigned int flags)
1568{
1569 int ret;
1570
1571 /*
1572 * Check ->nrbufs without the inode lock first. This function
1573 * is speculative anyways, so missing one is ok.
1574 */
1575 if (pipe->nrbufs < PIPE_BUFFERS)
1576 return 0;
1577
1578 ret = 0;
1579 mutex_lock(&pipe->inode->i_mutex);
1580
1581 while (pipe->nrbufs >= PIPE_BUFFERS) {
1582 if (!pipe->readers) {
1583 send_sig(SIGPIPE, current, 0);
1584 ret = -EPIPE;
1585 break;
1586 }
1587 if (flags & SPLICE_F_NONBLOCK) {
1588 ret = -EAGAIN;
1589 break;
1590 }
1591 if (signal_pending(current)) {
1592 ret = -ERESTARTSYS;
1593 break;
1594 }
1595 pipe->waiting_writers++;
1596 pipe_wait(pipe);
1597 pipe->waiting_writers--;
1598 }
1599
1600 mutex_unlock(&pipe->inode->i_mutex);
1601 return ret;
1602}
1603
1604/*
Jens Axboe70524492006-04-11 15:51:17 +02001605 * Link contents of ipipe to opipe.
1606 */
1607static int link_pipe(struct pipe_inode_info *ipipe,
1608 struct pipe_inode_info *opipe,
1609 size_t len, unsigned int flags)
1610{
1611 struct pipe_buffer *ibuf, *obuf;
Jens Axboeaadd06e2006-07-10 11:00:01 +02001612 int ret = 0, i = 0, nbuf;
Jens Axboe70524492006-04-11 15:51:17 +02001613
1614 /*
1615 * Potential ABBA deadlock, work around it by ordering lock
1616 * grabbing by inode address. Otherwise two different processes
1617 * could deadlock (one doing tee from A -> B, the other from B -> A).
1618 */
Mark Fasheh62752ee2006-10-17 10:31:38 +02001619 inode_double_lock(ipipe->inode, opipe->inode);
Jens Axboe70524492006-04-11 15:51:17 +02001620
Jens Axboeaadd06e2006-07-10 11:00:01 +02001621 do {
Jens Axboe70524492006-04-11 15:51:17 +02001622 if (!opipe->readers) {
1623 send_sig(SIGPIPE, current, 0);
1624 if (!ret)
1625 ret = -EPIPE;
1626 break;
1627 }
Jens Axboe70524492006-04-11 15:51:17 +02001628
1629 /*
Jens Axboeaadd06e2006-07-10 11:00:01 +02001630 * If we have iterated all input buffers or ran out of
1631 * output room, break.
Jens Axboe70524492006-04-11 15:51:17 +02001632 */
Jens Axboeaadd06e2006-07-10 11:00:01 +02001633 if (i >= ipipe->nrbufs || opipe->nrbufs >= PIPE_BUFFERS)
Jens Axboe70524492006-04-11 15:51:17 +02001634 break;
Jens Axboeaadd06e2006-07-10 11:00:01 +02001635
1636 ibuf = ipipe->bufs + ((ipipe->curbuf + i) & (PIPE_BUFFERS - 1));
1637 nbuf = (opipe->curbuf + opipe->nrbufs) & (PIPE_BUFFERS - 1);
1638
Jens Axboe2a272502006-04-19 15:56:40 +02001639 /*
Jens Axboeaadd06e2006-07-10 11:00:01 +02001640 * Get a reference to this pipe buffer,
1641 * so we can copy the contents over.
Jens Axboe2a272502006-04-19 15:56:40 +02001642 */
Jens Axboeaadd06e2006-07-10 11:00:01 +02001643 ibuf->ops->get(ipipe, ibuf);
Jens Axboe70524492006-04-11 15:51:17 +02001644
Jens Axboeaadd06e2006-07-10 11:00:01 +02001645 obuf = opipe->bufs + nbuf;
1646 *obuf = *ibuf;
Jens Axboe70524492006-04-11 15:51:17 +02001647
Jens Axboeaadd06e2006-07-10 11:00:01 +02001648 /*
1649 * Don't inherit the gift flag, we need to
1650 * prevent multiple steals of this page.
1651 */
1652 obuf->flags &= ~PIPE_BUF_FLAG_GIFT;
1653
1654 if (obuf->len > len)
1655 obuf->len = len;
1656
1657 opipe->nrbufs++;
1658 ret += obuf->len;
1659 len -= obuf->len;
1660 i++;
1661 } while (len);
Jens Axboe70524492006-04-11 15:51:17 +02001662
Mark Fasheh62752ee2006-10-17 10:31:38 +02001663 inode_double_unlock(ipipe->inode, opipe->inode);
Jens Axboe70524492006-04-11 15:51:17 +02001664
Jens Axboeaadd06e2006-07-10 11:00:01 +02001665 /*
1666 * If we put data in the output pipe, wakeup any potential readers.
1667 */
1668 if (ret > 0) {
Jens Axboe70524492006-04-11 15:51:17 +02001669 smp_mb();
1670 if (waitqueue_active(&opipe->wait))
1671 wake_up_interruptible(&opipe->wait);
1672 kill_fasync(&opipe->fasync_readers, SIGIO, POLL_IN);
1673 }
1674
1675 return ret;
1676}
1677
1678/*
1679 * This is a tee(1) implementation that works on pipes. It doesn't copy
1680 * any data, it simply references the 'in' pages on the 'out' pipe.
1681 * The 'flags' used are the SPLICE_F_* variants, currently the only
1682 * applicable one is SPLICE_F_NONBLOCK.
1683 */
1684static long do_tee(struct file *in, struct file *out, size_t len,
1685 unsigned int flags)
1686{
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001687 struct pipe_inode_info *ipipe = pipe_info(in->f_path.dentry->d_inode);
1688 struct pipe_inode_info *opipe = pipe_info(out->f_path.dentry->d_inode);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001689 int ret = -EINVAL;
Jens Axboe70524492006-04-11 15:51:17 +02001690
1691 /*
Jens Axboeaadd06e2006-07-10 11:00:01 +02001692 * Duplicate the contents of ipipe to opipe without actually
1693 * copying the data.
Jens Axboe70524492006-04-11 15:51:17 +02001694 */
Jens Axboeaadd06e2006-07-10 11:00:01 +02001695 if (ipipe && opipe && ipipe != opipe) {
1696 /*
1697 * Keep going, unless we encounter an error. The ipipe/opipe
1698 * ordering doesn't really matter.
1699 */
1700 ret = link_ipipe_prep(ipipe, flags);
1701 if (!ret) {
1702 ret = link_opipe_prep(opipe, flags);
1703 if (!ret) {
1704 ret = link_pipe(ipipe, opipe, len, flags);
1705 if (!ret && (flags & SPLICE_F_NONBLOCK))
1706 ret = -EAGAIN;
1707 }
1708 }
1709 }
Jens Axboe70524492006-04-11 15:51:17 +02001710
Jens Axboeaadd06e2006-07-10 11:00:01 +02001711 return ret;
Jens Axboe70524492006-04-11 15:51:17 +02001712}
1713
1714asmlinkage long sys_tee(int fdin, int fdout, size_t len, unsigned int flags)
1715{
1716 struct file *in;
1717 int error, fput_in;
1718
1719 if (unlikely(!len))
1720 return 0;
1721
1722 error = -EBADF;
1723 in = fget_light(fdin, &fput_in);
1724 if (in) {
1725 if (in->f_mode & FMODE_READ) {
1726 int fput_out;
1727 struct file *out = fget_light(fdout, &fput_out);
1728
1729 if (out) {
1730 if (out->f_mode & FMODE_WRITE)
1731 error = do_tee(in, out, len, flags);
1732 fput_light(out, fput_out);
1733 }
1734 }
1735 fput_light(in, fput_in);
1736 }
1737
1738 return error;
1739}