blob: ae50208e3e6c174e178b35d444de197a85341656 [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>
23#include <linux/pipe_fs_i.h>
24#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>
Jens Axboe5274f052006-03-30 15:15:30 +020031
Jens Axboe912d35f2006-04-26 10:59:21 +020032struct partial_page {
33 unsigned int offset;
34 unsigned int len;
35};
36
37/*
Jens Axboe00522fb2006-04-26 14:39:29 +020038 * Passed to splice_to_pipe
Jens Axboe912d35f2006-04-26 10:59:21 +020039 */
40struct splice_pipe_desc {
41 struct page **pages; /* page map */
42 struct partial_page *partial; /* pages[] may not be contig */
43 int nr_pages; /* number of pages in map */
44 unsigned int flags; /* splice flags */
Eric Dumazetd4c3cca2006-12-13 00:34:04 -080045 const struct pipe_buf_operations *ops;/* ops associated with output pipe */
Jens Axboe912d35f2006-04-26 10:59:21 +020046};
47
Jens Axboe83f91352006-04-02 23:05:09 +020048/*
49 * Attempt to steal a page from a pipe buffer. This should perhaps go into
50 * a vm helper function, it's already simplified quite a bit by the
51 * addition of remove_mapping(). If success is returned, the caller may
52 * attempt to reuse this page for another destination.
53 */
Jens Axboe76ad4d12006-05-03 10:41:33 +020054static int page_cache_pipe_buf_steal(struct pipe_inode_info *pipe,
Jens Axboe5abc97a2006-03-30 15:16:46 +020055 struct pipe_buffer *buf)
56{
57 struct page *page = buf->page;
Jens Axboe9e94cd42006-06-20 15:01:12 +020058 struct address_space *mapping;
Jens Axboe5abc97a2006-03-30 15:16:46 +020059
Jens Axboe9e0267c2006-04-19 15:57:31 +020060 lock_page(page);
61
Jens Axboe9e94cd42006-06-20 15:01:12 +020062 mapping = page_mapping(page);
63 if (mapping) {
64 WARN_ON(!PageUptodate(page));
Jens Axboe5abc97a2006-03-30 15:16:46 +020065
Jens Axboe9e94cd42006-06-20 15:01:12 +020066 /*
67 * At least for ext2 with nobh option, we need to wait on
68 * writeback completing on this page, since we'll remove it
69 * from the pagecache. Otherwise truncate wont wait on the
70 * page, allowing the disk blocks to be reused by someone else
71 * before we actually wrote our data to them. fs corruption
72 * ensues.
73 */
74 wait_on_page_writeback(page);
Jens Axboead8d6f02006-04-02 23:10:32 +020075
Jens Axboe9e94cd42006-06-20 15:01:12 +020076 if (PagePrivate(page))
Nick Piggin2ae88142006-10-28 10:38:23 -070077 try_to_release_page(page, GFP_KERNEL);
Jens Axboe4f6f0bd2006-04-02 23:04:46 +020078
Jens Axboe9e94cd42006-06-20 15:01:12 +020079 /*
80 * If we succeeded in removing the mapping, set LRU flag
81 * and return good.
82 */
83 if (remove_mapping(mapping, page)) {
84 buf->flags |= PIPE_BUF_FLAG_LRU;
85 return 0;
86 }
Jens Axboe9e0267c2006-04-19 15:57:31 +020087 }
Jens Axboe5abc97a2006-03-30 15:16:46 +020088
Jens Axboe9e94cd42006-06-20 15:01:12 +020089 /*
90 * Raced with truncate or failed to remove page from current
91 * address space, unlock and return failure.
92 */
93 unlock_page(page);
94 return 1;
Jens Axboe5abc97a2006-03-30 15:16:46 +020095}
96
Jens Axboe76ad4d12006-05-03 10:41:33 +020097static void page_cache_pipe_buf_release(struct pipe_inode_info *pipe,
Jens Axboe5274f052006-03-30 15:15:30 +020098 struct pipe_buffer *buf)
99{
100 page_cache_release(buf->page);
Jens Axboe14328732006-05-03 10:35:26 +0200101 buf->flags &= ~PIPE_BUF_FLAG_LRU;
Jens Axboe5274f052006-03-30 15:15:30 +0200102}
103
Jens Axboe76ad4d12006-05-03 10:41:33 +0200104static int page_cache_pipe_buf_pin(struct pipe_inode_info *pipe,
Jens Axboef84d7512006-05-01 19:59:03 +0200105 struct pipe_buffer *buf)
Jens Axboe5274f052006-03-30 15:15:30 +0200106{
107 struct page *page = buf->page;
Jens Axboe49d0b212006-04-10 09:04:41 +0200108 int err;
Jens Axboe5274f052006-03-30 15:15:30 +0200109
110 if (!PageUptodate(page)) {
Jens Axboe49d0b212006-04-10 09:04:41 +0200111 lock_page(page);
112
113 /*
114 * Page got truncated/unhashed. This will cause a 0-byte
Ingo Molnar73d62d82006-04-11 13:57:21 +0200115 * splice, if this is the first page.
Jens Axboe49d0b212006-04-10 09:04:41 +0200116 */
117 if (!page->mapping) {
118 err = -ENODATA;
119 goto error;
120 }
121
122 /*
Ingo Molnar73d62d82006-04-11 13:57:21 +0200123 * Uh oh, read-error from disk.
Jens Axboe49d0b212006-04-10 09:04:41 +0200124 */
125 if (!PageUptodate(page)) {
126 err = -EIO;
127 goto error;
128 }
129
130 /*
Jens Axboef84d7512006-05-01 19:59:03 +0200131 * Page is ok afterall, we are done.
Jens Axboe49d0b212006-04-10 09:04:41 +0200132 */
Jens Axboe5274f052006-03-30 15:15:30 +0200133 unlock_page(page);
Jens Axboe5274f052006-03-30 15:15:30 +0200134 }
135
Jens Axboef84d7512006-05-01 19:59:03 +0200136 return 0;
Jens Axboe49d0b212006-04-10 09:04:41 +0200137error:
138 unlock_page(page);
Jens Axboef84d7512006-05-01 19:59:03 +0200139 return err;
Jens Axboe70524492006-04-11 15:51:17 +0200140}
141
Eric Dumazetd4c3cca2006-12-13 00:34:04 -0800142static const struct pipe_buf_operations page_cache_pipe_buf_ops = {
Jens Axboe5274f052006-03-30 15:15:30 +0200143 .can_merge = 0,
Jens Axboef84d7512006-05-01 19:59:03 +0200144 .map = generic_pipe_buf_map,
145 .unmap = generic_pipe_buf_unmap,
146 .pin = page_cache_pipe_buf_pin,
Jens Axboe5274f052006-03-30 15:15:30 +0200147 .release = page_cache_pipe_buf_release,
Jens Axboe5abc97a2006-03-30 15:16:46 +0200148 .steal = page_cache_pipe_buf_steal,
Jens Axboef84d7512006-05-01 19:59:03 +0200149 .get = generic_pipe_buf_get,
Jens Axboe5274f052006-03-30 15:15:30 +0200150};
151
Jens Axboe912d35f2006-04-26 10:59:21 +0200152static int user_page_pipe_buf_steal(struct pipe_inode_info *pipe,
153 struct pipe_buffer *buf)
154{
Jens Axboe7afa6fd2006-05-01 20:02:33 +0200155 if (!(buf->flags & PIPE_BUF_FLAG_GIFT))
156 return 1;
157
Jens Axboe14328732006-05-03 10:35:26 +0200158 buf->flags |= PIPE_BUF_FLAG_LRU;
Jens Axboe330ab712006-05-02 15:29:57 +0200159 return generic_pipe_buf_steal(pipe, buf);
Jens Axboe912d35f2006-04-26 10:59:21 +0200160}
161
Eric Dumazetd4c3cca2006-12-13 00:34:04 -0800162static const struct pipe_buf_operations user_page_pipe_buf_ops = {
Jens Axboe912d35f2006-04-26 10:59:21 +0200163 .can_merge = 0,
Jens Axboef84d7512006-05-01 19:59:03 +0200164 .map = generic_pipe_buf_map,
165 .unmap = generic_pipe_buf_unmap,
166 .pin = generic_pipe_buf_pin,
Jens Axboe912d35f2006-04-26 10:59:21 +0200167 .release = page_cache_pipe_buf_release,
168 .steal = user_page_pipe_buf_steal,
Jens Axboef84d7512006-05-01 19:59:03 +0200169 .get = generic_pipe_buf_get,
Jens Axboe912d35f2006-04-26 10:59:21 +0200170};
171
Jens Axboe83f91352006-04-02 23:05:09 +0200172/*
173 * Pipe output worker. This sets up our pipe format with the page cache
174 * pipe buffer operations. Otherwise very similar to the regular pipe_writev().
175 */
Jens Axboe00522fb2006-04-26 14:39:29 +0200176static ssize_t splice_to_pipe(struct pipe_inode_info *pipe,
177 struct splice_pipe_desc *spd)
Jens Axboe5274f052006-03-30 15:15:30 +0200178{
Jens Axboe912d35f2006-04-26 10:59:21 +0200179 int ret, do_wakeup, page_nr;
Jens Axboe5274f052006-03-30 15:15:30 +0200180
181 ret = 0;
182 do_wakeup = 0;
Jens Axboe912d35f2006-04-26 10:59:21 +0200183 page_nr = 0;
Jens Axboe5274f052006-03-30 15:15:30 +0200184
Ingo Molnar3a326a22006-04-10 15:18:35 +0200185 if (pipe->inode)
186 mutex_lock(&pipe->inode->i_mutex);
Jens Axboe5274f052006-03-30 15:15:30 +0200187
Jens Axboe5274f052006-03-30 15:15:30 +0200188 for (;;) {
Ingo Molnar3a326a22006-04-10 15:18:35 +0200189 if (!pipe->readers) {
Jens Axboe5274f052006-03-30 15:15:30 +0200190 send_sig(SIGPIPE, current, 0);
191 if (!ret)
192 ret = -EPIPE;
193 break;
194 }
195
Jens Axboe6f767b02006-04-11 13:53:56 +0200196 if (pipe->nrbufs < PIPE_BUFFERS) {
197 int newbuf = (pipe->curbuf + pipe->nrbufs) & (PIPE_BUFFERS - 1);
Ingo Molnar3a326a22006-04-10 15:18:35 +0200198 struct pipe_buffer *buf = pipe->bufs + newbuf;
Jens Axboe5274f052006-03-30 15:15:30 +0200199
Jens Axboe912d35f2006-04-26 10:59:21 +0200200 buf->page = spd->pages[page_nr];
201 buf->offset = spd->partial[page_nr].offset;
202 buf->len = spd->partial[page_nr].len;
203 buf->ops = spd->ops;
Jens Axboe7afa6fd2006-05-01 20:02:33 +0200204 if (spd->flags & SPLICE_F_GIFT)
205 buf->flags |= PIPE_BUF_FLAG_GIFT;
206
Jens Axboe6f767b02006-04-11 13:53:56 +0200207 pipe->nrbufs++;
Jens Axboe912d35f2006-04-26 10:59:21 +0200208 page_nr++;
209 ret += buf->len;
210
Jens Axboe6f767b02006-04-11 13:53:56 +0200211 if (pipe->inode)
212 do_wakeup = 1;
Jens Axboe5274f052006-03-30 15:15:30 +0200213
Jens Axboe912d35f2006-04-26 10:59:21 +0200214 if (!--spd->nr_pages)
Jens Axboe5274f052006-03-30 15:15:30 +0200215 break;
Jens Axboe6f767b02006-04-11 13:53:56 +0200216 if (pipe->nrbufs < PIPE_BUFFERS)
Jens Axboe5274f052006-03-30 15:15:30 +0200217 continue;
218
219 break;
220 }
221
Jens Axboe912d35f2006-04-26 10:59:21 +0200222 if (spd->flags & SPLICE_F_NONBLOCK) {
Linus Torvalds29e35092006-04-02 12:46:35 -0700223 if (!ret)
224 ret = -EAGAIN;
225 break;
226 }
227
Jens Axboe5274f052006-03-30 15:15:30 +0200228 if (signal_pending(current)) {
229 if (!ret)
230 ret = -ERESTARTSYS;
231 break;
232 }
233
234 if (do_wakeup) {
Jens Axboec0bd1f62006-04-10 09:03:32 +0200235 smp_mb();
Ingo Molnar3a326a22006-04-10 15:18:35 +0200236 if (waitqueue_active(&pipe->wait))
237 wake_up_interruptible_sync(&pipe->wait);
238 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
Jens Axboe5274f052006-03-30 15:15:30 +0200239 do_wakeup = 0;
240 }
241
Ingo Molnar3a326a22006-04-10 15:18:35 +0200242 pipe->waiting_writers++;
243 pipe_wait(pipe);
244 pipe->waiting_writers--;
Jens Axboe5274f052006-03-30 15:15:30 +0200245 }
246
Ingo Molnar3a326a22006-04-10 15:18:35 +0200247 if (pipe->inode)
248 mutex_unlock(&pipe->inode->i_mutex);
Jens Axboe5274f052006-03-30 15:15:30 +0200249
250 if (do_wakeup) {
Jens Axboec0bd1f62006-04-10 09:03:32 +0200251 smp_mb();
Ingo Molnar3a326a22006-04-10 15:18:35 +0200252 if (waitqueue_active(&pipe->wait))
253 wake_up_interruptible(&pipe->wait);
254 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
Jens Axboe5274f052006-03-30 15:15:30 +0200255 }
256
Jens Axboe912d35f2006-04-26 10:59:21 +0200257 while (page_nr < spd->nr_pages)
258 page_cache_release(spd->pages[page_nr++]);
Jens Axboe5274f052006-03-30 15:15:30 +0200259
260 return ret;
261}
262
Ingo Molnar3a326a22006-04-10 15:18:35 +0200263static int
Jens Axboecbb7e572006-04-11 14:57:50 +0200264__generic_file_splice_read(struct file *in, loff_t *ppos,
265 struct pipe_inode_info *pipe, size_t len,
266 unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +0200267{
268 struct address_space *mapping = in->f_mapping;
Jens Axboe912d35f2006-04-26 10:59:21 +0200269 unsigned int loff, nr_pages;
Jens Axboe16c523d2006-04-10 09:03:58 +0200270 struct page *pages[PIPE_BUFFERS];
Jens Axboe912d35f2006-04-26 10:59:21 +0200271 struct partial_page partial[PIPE_BUFFERS];
Jens Axboe5274f052006-03-30 15:15:30 +0200272 struct page *page;
Jens Axboe91ad66e2006-04-19 15:55:10 +0200273 pgoff_t index, end_index;
274 loff_t isize;
Jens Axboe912d35f2006-04-26 10:59:21 +0200275 size_t total_len;
Jens Axboeeb207962006-04-27 11:05:22 +0200276 int error, page_nr;
Jens Axboe912d35f2006-04-26 10:59:21 +0200277 struct splice_pipe_desc spd = {
278 .pages = pages,
279 .partial = partial,
280 .flags = flags,
281 .ops = &page_cache_pipe_buf_ops,
282 };
Jens Axboe5274f052006-03-30 15:15:30 +0200283
Jens Axboecbb7e572006-04-11 14:57:50 +0200284 index = *ppos >> PAGE_CACHE_SHIFT;
Jens Axboe912d35f2006-04-26 10:59:21 +0200285 loff = *ppos & ~PAGE_CACHE_MASK;
286 nr_pages = (len + loff + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
Jens Axboe5274f052006-03-30 15:15:30 +0200287
288 if (nr_pages > PIPE_BUFFERS)
289 nr_pages = PIPE_BUFFERS;
290
291 /*
Ingo Molnar73d62d82006-04-11 13:57:21 +0200292 * Initiate read-ahead on this page range. however, don't call into
Jens Axboe0b749ce2006-04-10 09:05:04 +0200293 * read-ahead if this is a non-zero offset (we are likely doing small
294 * chunk splice and the page is already there) for a single page.
Jens Axboe5274f052006-03-30 15:15:30 +0200295 */
Jens Axboeeb645a22006-04-27 08:44:27 +0200296 if (!loff || nr_pages > 1)
297 page_cache_readahead(mapping, &in->f_ra, in, index, nr_pages);
Jens Axboe5274f052006-03-30 15:15:30 +0200298
299 /*
Ingo Molnar73d62d82006-04-11 13:57:21 +0200300 * Now fill in the holes:
Jens Axboe5274f052006-03-30 15:15:30 +0200301 */
Jens Axboe7480a902006-04-11 13:52:47 +0200302 error = 0;
Jens Axboe912d35f2006-04-26 10:59:21 +0200303 total_len = 0;
Jens Axboeeb207962006-04-27 11:05:22 +0200304
305 /*
306 * Lookup the (hopefully) full range of pages we need.
307 */
308 spd.nr_pages = find_get_pages_contig(mapping, index, nr_pages, pages);
309
310 /*
311 * If find_get_pages_contig() returned fewer pages than we needed,
312 * allocate the rest.
313 */
314 index += spd.nr_pages;
315 while (spd.nr_pages < nr_pages) {
316 /*
317 * Page could be there, find_get_pages_contig() breaks on
318 * the first hole.
319 */
320 page = find_get_page(mapping, index);
321 if (!page) {
322 /*
Jens Axboee27dedd2006-05-01 19:59:54 +0200323 * Make sure the read-ahead engine is notified
324 * about this failure.
325 */
326 handle_ra_miss(mapping, &in->f_ra, index);
327
328 /*
Jens Axboeeb207962006-04-27 11:05:22 +0200329 * page didn't exist, allocate one.
330 */
331 page = page_cache_alloc_cold(mapping);
332 if (!page)
333 break;
334
335 error = add_to_page_cache_lru(page, mapping, index,
Nick Piggin2ae88142006-10-28 10:38:23 -0700336 GFP_KERNEL);
Jens Axboeeb207962006-04-27 11:05:22 +0200337 if (unlikely(error)) {
338 page_cache_release(page);
Jens Axboea0548872006-05-03 10:58:22 +0200339 if (error == -EEXIST)
340 continue;
Jens Axboeeb207962006-04-27 11:05:22 +0200341 break;
342 }
343 /*
344 * add_to_page_cache() locks the page, unlock it
345 * to avoid convoluting the logic below even more.
346 */
347 unlock_page(page);
348 }
349
350 pages[spd.nr_pages++] = page;
351 index++;
352 }
353
354 /*
355 * Now loop over the map and see if we need to start IO on any
356 * pages, fill in the partial map, etc.
357 */
358 index = *ppos >> PAGE_CACHE_SHIFT;
359 nr_pages = spd.nr_pages;
360 spd.nr_pages = 0;
361 for (page_nr = 0; page_nr < nr_pages; page_nr++) {
Jens Axboe82aa5d62006-04-20 13:05:48 +0200362 unsigned int this_len;
363
364 if (!len)
365 break;
366
367 /*
368 * this_len is the max we'll use from this page
369 */
Andrew Mortonba5f5d92006-04-25 15:33:34 +0200370 this_len = min_t(unsigned long, len, PAGE_CACHE_SIZE - loff);
Jens Axboeeb207962006-04-27 11:05:22 +0200371 page = pages[page_nr];
Jens Axboe7480a902006-04-11 13:52:47 +0200372
373 /*
374 * If the page isn't uptodate, we may need to start io on it
375 */
376 if (!PageUptodate(page)) {
Jens Axboec4f895c2006-04-19 15:56:12 +0200377 /*
378 * If in nonblock mode then dont block on waiting
379 * for an in-flight io page
380 */
381 if (flags & SPLICE_F_NONBLOCK)
382 break;
383
Jens Axboe7480a902006-04-11 13:52:47 +0200384 lock_page(page);
385
386 /*
387 * page was truncated, stop here. if this isn't the
388 * first page, we'll just complete what we already
389 * added
390 */
391 if (!page->mapping) {
392 unlock_page(page);
Jens Axboe7480a902006-04-11 13:52:47 +0200393 break;
394 }
395 /*
396 * page was already under io and is now done, great
397 */
398 if (PageUptodate(page)) {
399 unlock_page(page);
400 goto fill_it;
401 }
402
Jens Axboe7480a902006-04-11 13:52:47 +0200403 /*
404 * need to read in the page
405 */
406 error = mapping->a_ops->readpage(in, page);
Jens Axboe7480a902006-04-11 13:52:47 +0200407 if (unlikely(error)) {
Jens Axboeeb207962006-04-27 11:05:22 +0200408 /*
409 * We really should re-lookup the page here,
410 * but it complicates things a lot. Instead
411 * lets just do what we already stored, and
412 * we'll get it the next time we are called.
413 */
Jens Axboe7480a902006-04-11 13:52:47 +0200414 if (error == AOP_TRUNCATED_PAGE)
Jens Axboeeb207962006-04-27 11:05:22 +0200415 error = 0;
416
Jens Axboe7480a902006-04-11 13:52:47 +0200417 break;
418 }
Jens Axboe91ad66e2006-04-19 15:55:10 +0200419
420 /*
421 * i_size must be checked after ->readpage().
422 */
423 isize = i_size_read(mapping->host);
424 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
Jens Axboeeb207962006-04-27 11:05:22 +0200425 if (unlikely(!isize || index > end_index))
Jens Axboe91ad66e2006-04-19 15:55:10 +0200426 break;
Jens Axboe91ad66e2006-04-19 15:55:10 +0200427
428 /*
429 * if this is the last page, see if we need to shrink
430 * the length and stop
431 */
432 if (end_index == index) {
433 loff = PAGE_CACHE_SIZE - (isize & ~PAGE_CACHE_MASK);
Jens Axboeeb207962006-04-27 11:05:22 +0200434 if (total_len + loff > isize)
Jens Axboe91ad66e2006-04-19 15:55:10 +0200435 break;
Jens Axboe91ad66e2006-04-19 15:55:10 +0200436 /*
437 * force quit after adding this page
438 */
Jens Axboeeb207962006-04-27 11:05:22 +0200439 len = this_len;
Jens Axboe82aa5d62006-04-20 13:05:48 +0200440 this_len = min(this_len, loff);
Jens Axboe912d35f2006-04-26 10:59:21 +0200441 loff = 0;
Jens Axboe91ad66e2006-04-19 15:55:10 +0200442 }
Jens Axboe7480a902006-04-11 13:52:47 +0200443 }
444fill_it:
Jens Axboeeb207962006-04-27 11:05:22 +0200445 partial[page_nr].offset = loff;
446 partial[page_nr].len = this_len;
Jens Axboe82aa5d62006-04-20 13:05:48 +0200447 len -= this_len;
Jens Axboe912d35f2006-04-26 10:59:21 +0200448 total_len += this_len;
Jens Axboe91ad66e2006-04-19 15:55:10 +0200449 loff = 0;
Jens Axboeeb207962006-04-27 11:05:22 +0200450 spd.nr_pages++;
451 index++;
Jens Axboe5274f052006-03-30 15:15:30 +0200452 }
453
Jens Axboeeb207962006-04-27 11:05:22 +0200454 /*
455 * Release any pages at the end, if we quit early. 'i' is how far
456 * we got, 'nr_pages' is how many pages are in the map.
457 */
458 while (page_nr < nr_pages)
459 page_cache_release(pages[page_nr++]);
460
Jens Axboe912d35f2006-04-26 10:59:21 +0200461 if (spd.nr_pages)
Jens Axboe00522fb2006-04-26 14:39:29 +0200462 return splice_to_pipe(pipe, &spd);
Jens Axboe5274f052006-03-30 15:15:30 +0200463
Jens Axboe7480a902006-04-11 13:52:47 +0200464 return error;
Jens Axboe5274f052006-03-30 15:15:30 +0200465}
466
Jens Axboe83f91352006-04-02 23:05:09 +0200467/**
468 * generic_file_splice_read - splice data from file to a pipe
469 * @in: file to splice from
470 * @pipe: pipe to splice to
471 * @len: number of bytes to splice
472 * @flags: splice modifier flags
473 *
474 * Will read pages from given file and fill them into a pipe.
Jens Axboe83f91352006-04-02 23:05:09 +0200475 */
Jens Axboecbb7e572006-04-11 14:57:50 +0200476ssize_t generic_file_splice_read(struct file *in, loff_t *ppos,
477 struct pipe_inode_info *pipe, size_t len,
478 unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +0200479{
480 ssize_t spliced;
481 int ret;
482
483 ret = 0;
484 spliced = 0;
Ingo Molnar3a326a22006-04-10 15:18:35 +0200485
Jens Axboe5274f052006-03-30 15:15:30 +0200486 while (len) {
Jens Axboecbb7e572006-04-11 14:57:50 +0200487 ret = __generic_file_splice_read(in, ppos, pipe, len, flags);
Jens Axboe5274f052006-03-30 15:15:30 +0200488
Jens Axboec4f895c2006-04-19 15:56:12 +0200489 if (ret < 0)
Jens Axboe5274f052006-03-30 15:15:30 +0200490 break;
Jens Axboec4f895c2006-04-19 15:56:12 +0200491 else if (!ret) {
492 if (spliced)
493 break;
494 if (flags & SPLICE_F_NONBLOCK) {
495 ret = -EAGAIN;
496 break;
497 }
498 }
Jens Axboe5274f052006-03-30 15:15:30 +0200499
Jens Axboecbb7e572006-04-11 14:57:50 +0200500 *ppos += ret;
Jens Axboe5274f052006-03-30 15:15:30 +0200501 len -= ret;
502 spliced += ret;
503 }
504
505 if (spliced)
506 return spliced;
507
508 return ret;
509}
510
Jens Axboe059a8f32006-04-02 23:06:05 +0200511EXPORT_SYMBOL(generic_file_splice_read);
512
Jens Axboe5274f052006-03-30 15:15:30 +0200513/*
Jens Axboe4f6f0bd2006-04-02 23:04:46 +0200514 * Send 'sd->len' bytes to socket from 'sd->file' at position 'sd->pos'
Jens Axboe016b6612006-04-25 15:42:00 +0200515 * using sendpage(). Return the number of bytes sent.
Jens Axboe5274f052006-03-30 15:15:30 +0200516 */
Jens Axboe76ad4d12006-05-03 10:41:33 +0200517static int pipe_to_sendpage(struct pipe_inode_info *pipe,
Jens Axboe5274f052006-03-30 15:15:30 +0200518 struct pipe_buffer *buf, struct splice_desc *sd)
519{
520 struct file *file = sd->file;
521 loff_t pos = sd->pos;
Jens Axboef84d7512006-05-01 19:59:03 +0200522 int ret, more;
Jens Axboe5274f052006-03-30 15:15:30 +0200523
Jens Axboe76ad4d12006-05-03 10:41:33 +0200524 ret = buf->ops->pin(pipe, buf);
Jens Axboef84d7512006-05-01 19:59:03 +0200525 if (!ret) {
526 more = (sd->flags & SPLICE_F_MORE) || sd->len < sd->total_len;
Jens Axboe5274f052006-03-30 15:15:30 +0200527
Jens Axboef84d7512006-05-01 19:59:03 +0200528 ret = file->f_op->sendpage(file, buf->page, buf->offset,
529 sd->len, &pos, more);
530 }
Jens Axboe5274f052006-03-30 15:15:30 +0200531
Jens Axboe016b6612006-04-25 15:42:00 +0200532 return ret;
Jens Axboe5274f052006-03-30 15:15:30 +0200533}
534
535/*
536 * This is a little more tricky than the file -> pipe splicing. There are
537 * basically three cases:
538 *
539 * - Destination page already exists in the address space and there
540 * are users of it. For that case we have no other option that
541 * copying the data. Tough luck.
542 * - Destination page already exists in the address space, but there
543 * are no users of it. Make sure it's uptodate, then drop it. Fall
544 * through to last case.
545 * - Destination page does not exist, we can add the pipe page to
546 * the page cache and avoid the copy.
547 *
Jens Axboe83f91352006-04-02 23:05:09 +0200548 * If asked to move pages to the output file (SPLICE_F_MOVE is set in
549 * sd->flags), we attempt to migrate pages from the pipe to the output
550 * file address space page cache. This is possible if no one else has
551 * the pipe page referenced outside of the pipe and page cache. If
552 * SPLICE_F_MOVE isn't set, or we cannot move the page, we simply create
553 * a new page in the output file page cache and fill/dirty that.
Jens Axboe5274f052006-03-30 15:15:30 +0200554 */
Jens Axboe76ad4d12006-05-03 10:41:33 +0200555static int pipe_to_file(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
Jens Axboe5274f052006-03-30 15:15:30 +0200556 struct splice_desc *sd)
557{
558 struct file *file = sd->file;
559 struct address_space *mapping = file->f_mapping;
Jens Axboe016b6612006-04-25 15:42:00 +0200560 unsigned int offset, this_len;
Jens Axboe5274f052006-03-30 15:15:30 +0200561 struct page *page;
Jens Axboe5274f052006-03-30 15:15:30 +0200562 pgoff_t index;
Jens Axboe3e7ee3e2006-04-02 23:11:04 +0200563 int ret;
Jens Axboe5274f052006-03-30 15:15:30 +0200564
565 /*
Jens Axboe49d0b212006-04-10 09:04:41 +0200566 * make sure the data in this buffer is uptodate
Jens Axboe5274f052006-03-30 15:15:30 +0200567 */
Jens Axboe76ad4d12006-05-03 10:41:33 +0200568 ret = buf->ops->pin(pipe, buf);
Jens Axboef84d7512006-05-01 19:59:03 +0200569 if (unlikely(ret))
570 return ret;
Jens Axboe5274f052006-03-30 15:15:30 +0200571
572 index = sd->pos >> PAGE_CACHE_SHIFT;
573 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
Jens Axboe5abc97a2006-03-30 15:16:46 +0200579find_page:
Nick Piggin485ddb42007-03-27 08:55:08 +0200580 page = find_lock_page(mapping, index);
581 if (!page) {
582 ret = -ENOMEM;
583 page = page_cache_alloc_cold(mapping);
584 if (unlikely(!page))
585 goto out_ret;
Jens Axboe9e0267c2006-04-19 15:57:31 +0200586
Nick Piggin485ddb42007-03-27 08:55:08 +0200587 /*
588 * This will also lock the page
589 */
590 ret = add_to_page_cache_lru(page, mapping, index,
591 GFP_KERNEL);
592 if (unlikely(ret))
593 goto out;
594 }
595
Jens Axboe016b6612006-04-25 15:42:00 +0200596 ret = mapping->a_ops->prepare_write(file, page, offset, offset+this_len);
Jens Axboebfc4ee32006-05-03 10:35:10 +0200597 if (unlikely(ret)) {
598 loff_t isize = i_size_read(mapping->host);
599
600 if (ret != AOP_TRUNCATED_PAGE)
601 unlock_page(page);
Jens Axboe4f6f0bd2006-04-02 23:04:46 +0200602 page_cache_release(page);
Jens Axboebfc4ee32006-05-03 10:35:10 +0200603 if (ret == AOP_TRUNCATED_PAGE)
604 goto find_page;
605
606 /*
607 * prepare_write() may have instantiated a few blocks
608 * outside i_size. Trim these off again.
609 */
610 if (sd->pos + this_len > isize)
611 vmtruncate(mapping->host, isize);
612
Jens Axboee6e80f22006-10-11 10:03:09 +0200613 goto out_ret;
Jens Axboebfc4ee32006-05-03 10:35:10 +0200614 }
Jens Axboe5274f052006-03-30 15:15:30 +0200615
Jens Axboe0568b402006-05-01 19:50:48 +0200616 if (buf->page != page) {
Jens Axboef84d7512006-05-01 19:59:03 +0200617 /*
618 * Careful, ->map() uses KM_USER0!
619 */
Jens Axboe76ad4d12006-05-03 10:41:33 +0200620 char *src = buf->ops->map(pipe, buf, 1);
Jens Axboef84d7512006-05-01 19:59:03 +0200621 char *dst = kmap_atomic(page, KM_USER1);
Jens Axboe5abc97a2006-03-30 15:16:46 +0200622
Jens Axboe016b6612006-04-25 15:42:00 +0200623 memcpy(dst + offset, src + buf->offset, this_len);
Jens Axboe5abc97a2006-03-30 15:16:46 +0200624 flush_dcache_page(page);
Jens Axboef84d7512006-05-01 19:59:03 +0200625 kunmap_atomic(dst, KM_USER1);
Jens Axboe76ad4d12006-05-03 10:41:33 +0200626 buf->ops->unmap(pipe, buf, src);
Jens Axboe5abc97a2006-03-30 15:16:46 +0200627 }
Jens Axboe5274f052006-03-30 15:15:30 +0200628
Jens Axboe016b6612006-04-25 15:42:00 +0200629 ret = mapping->a_ops->commit_write(file, page, offset, offset+this_len);
Jens Axboe0568b402006-05-01 19:50:48 +0200630 if (!ret) {
631 /*
632 * Return the number of bytes written and mark page as
633 * accessed, we are now done!
634 */
635 ret = this_len;
636 mark_page_accessed(page);
637 balance_dirty_pages_ratelimited(mapping);
638 } else if (ret == AOP_TRUNCATED_PAGE) {
Jens Axboe4f6f0bd2006-04-02 23:04:46 +0200639 page_cache_release(page);
640 goto find_page;
Jens Axboe0568b402006-05-01 19:50:48 +0200641 }
Jens Axboe5274f052006-03-30 15:15:30 +0200642out:
Jens Axboe0568b402006-05-01 19:50:48 +0200643 page_cache_release(page);
Jens Axboe9e0267c2006-04-19 15:57:31 +0200644 unlock_page(page);
Jens Axboee6e80f22006-10-11 10:03:09 +0200645out_ret:
Jens Axboe5274f052006-03-30 15:15:30 +0200646 return ret;
647}
648
Jens Axboe83f91352006-04-02 23:05:09 +0200649/*
650 * Pipe input worker. Most of this logic works like a regular pipe, the
651 * key here is the 'actor' worker passed in that actually moves the data
652 * to the wanted destination. See pipe_to_file/pipe_to_sendpage above.
653 */
Mark Fasheh6da61802006-10-17 18:43:07 +0200654static ssize_t __splice_from_pipe(struct pipe_inode_info *pipe,
655 struct file *out, loff_t *ppos, size_t len,
656 unsigned int flags, splice_actor *actor)
Jens Axboe5274f052006-03-30 15:15:30 +0200657{
Jens Axboe5274f052006-03-30 15:15:30 +0200658 int ret, do_wakeup, err;
659 struct splice_desc sd;
660
661 ret = 0;
662 do_wakeup = 0;
663
664 sd.total_len = len;
665 sd.flags = flags;
666 sd.file = out;
Jens Axboecbb7e572006-04-11 14:57:50 +0200667 sd.pos = *ppos;
Jens Axboe5274f052006-03-30 15:15:30 +0200668
Jens Axboe5274f052006-03-30 15:15:30 +0200669 for (;;) {
Jens Axboe6f767b02006-04-11 13:53:56 +0200670 if (pipe->nrbufs) {
671 struct pipe_buffer *buf = pipe->bufs + pipe->curbuf;
Eric Dumazetd4c3cca2006-12-13 00:34:04 -0800672 const struct pipe_buf_operations *ops = buf->ops;
Jens Axboe5274f052006-03-30 15:15:30 +0200673
674 sd.len = buf->len;
675 if (sd.len > sd.total_len)
676 sd.len = sd.total_len;
677
Ingo Molnar3a326a22006-04-10 15:18:35 +0200678 err = actor(pipe, buf, &sd);
Jens Axboe016b6612006-04-25 15:42:00 +0200679 if (err <= 0) {
Jens Axboe5274f052006-03-30 15:15:30 +0200680 if (!ret && err != -ENODATA)
681 ret = err;
682
683 break;
684 }
685
Jens Axboe016b6612006-04-25 15:42:00 +0200686 ret += err;
687 buf->offset += err;
688 buf->len -= err;
689
690 sd.len -= err;
691 sd.pos += err;
692 sd.total_len -= err;
693 if (sd.len)
694 continue;
Ingo Molnar73d62d82006-04-11 13:57:21 +0200695
Jens Axboe5274f052006-03-30 15:15:30 +0200696 if (!buf->len) {
697 buf->ops = NULL;
Ingo Molnar3a326a22006-04-10 15:18:35 +0200698 ops->release(pipe, buf);
Jens Axboe6f767b02006-04-11 13:53:56 +0200699 pipe->curbuf = (pipe->curbuf + 1) & (PIPE_BUFFERS - 1);
700 pipe->nrbufs--;
701 if (pipe->inode)
702 do_wakeup = 1;
Jens Axboe5274f052006-03-30 15:15:30 +0200703 }
704
Jens Axboe5274f052006-03-30 15:15:30 +0200705 if (!sd.total_len)
706 break;
707 }
708
Jens Axboe6f767b02006-04-11 13:53:56 +0200709 if (pipe->nrbufs)
Jens Axboe5274f052006-03-30 15:15:30 +0200710 continue;
Ingo Molnar3a326a22006-04-10 15:18:35 +0200711 if (!pipe->writers)
Jens Axboe5274f052006-03-30 15:15:30 +0200712 break;
Ingo Molnar3a326a22006-04-10 15:18:35 +0200713 if (!pipe->waiting_writers) {
Jens Axboe5274f052006-03-30 15:15:30 +0200714 if (ret)
715 break;
716 }
717
Linus Torvalds29e35092006-04-02 12:46:35 -0700718 if (flags & SPLICE_F_NONBLOCK) {
719 if (!ret)
720 ret = -EAGAIN;
721 break;
722 }
723
Jens Axboe5274f052006-03-30 15:15:30 +0200724 if (signal_pending(current)) {
725 if (!ret)
726 ret = -ERESTARTSYS;
727 break;
728 }
729
730 if (do_wakeup) {
Jens Axboec0bd1f62006-04-10 09:03:32 +0200731 smp_mb();
Ingo Molnar3a326a22006-04-10 15:18:35 +0200732 if (waitqueue_active(&pipe->wait))
733 wake_up_interruptible_sync(&pipe->wait);
734 kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
Jens Axboe5274f052006-03-30 15:15:30 +0200735 do_wakeup = 0;
736 }
737
Ingo Molnar3a326a22006-04-10 15:18:35 +0200738 pipe_wait(pipe);
Jens Axboe5274f052006-03-30 15:15:30 +0200739 }
740
Jens Axboe5274f052006-03-30 15:15:30 +0200741 if (do_wakeup) {
Jens Axboec0bd1f62006-04-10 09:03:32 +0200742 smp_mb();
Ingo Molnar3a326a22006-04-10 15:18:35 +0200743 if (waitqueue_active(&pipe->wait))
744 wake_up_interruptible(&pipe->wait);
745 kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
Jens Axboe5274f052006-03-30 15:15:30 +0200746 }
747
Jens Axboe5274f052006-03-30 15:15:30 +0200748 return ret;
Jens Axboe5274f052006-03-30 15:15:30 +0200749}
750
Mark Fasheh6da61802006-10-17 18:43:07 +0200751ssize_t splice_from_pipe(struct pipe_inode_info *pipe, struct file *out,
752 loff_t *ppos, size_t len, unsigned int flags,
753 splice_actor *actor)
754{
755 ssize_t ret;
756 struct inode *inode = out->f_mapping->host;
757
758 /*
759 * The actor worker might be calling ->prepare_write and
760 * ->commit_write. Most of the time, these expect i_mutex to
761 * be held. Since this may result in an ABBA deadlock with
762 * pipe->inode, we have to order lock acquiry here.
763 */
764 inode_double_lock(inode, pipe->inode);
765 ret = __splice_from_pipe(pipe, out, ppos, len, flags, actor);
766 inode_double_unlock(inode, pipe->inode);
767
768 return ret;
769}
770
771/**
772 * generic_file_splice_write_nolock - generic_file_splice_write without mutexes
773 * @pipe: pipe info
774 * @out: file to write to
775 * @len: number of bytes to splice
776 * @flags: splice modifier flags
777 *
778 * Will either move or copy pages (determined by @flags options) from
779 * the given pipe inode to the given file. The caller is responsible
780 * for acquiring i_mutex on both inodes.
781 *
782 */
783ssize_t
784generic_file_splice_write_nolock(struct pipe_inode_info *pipe, struct file *out,
785 loff_t *ppos, size_t len, unsigned int flags)
786{
787 struct address_space *mapping = out->f_mapping;
788 struct inode *inode = mapping->host;
789 ssize_t ret;
790 int err;
791
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800792 err = remove_suid(out->f_path.dentry);
Jens Axboe8c34e2d2006-10-17 19:43:22 +0200793 if (unlikely(err))
794 return err;
795
Mark Fasheh6da61802006-10-17 18:43:07 +0200796 ret = __splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_file);
797 if (ret > 0) {
798 *ppos += ret;
799
800 /*
801 * If file or inode is SYNC and we actually wrote some data,
802 * sync it.
803 */
804 if (unlikely((out->f_flags & O_SYNC) || IS_SYNC(inode))) {
805 err = generic_osync_inode(inode, mapping,
806 OSYNC_METADATA|OSYNC_DATA);
807
808 if (err)
809 ret = err;
810 }
811 }
812
813 return ret;
814}
815
816EXPORT_SYMBOL(generic_file_splice_write_nolock);
817
Jens Axboe83f91352006-04-02 23:05:09 +0200818/**
819 * generic_file_splice_write - splice data from a pipe to a file
Ingo Molnar3a326a22006-04-10 15:18:35 +0200820 * @pipe: pipe info
Jens Axboe83f91352006-04-02 23:05:09 +0200821 * @out: file to write to
822 * @len: number of bytes to splice
823 * @flags: splice modifier flags
824 *
825 * Will either move or copy pages (determined by @flags options) from
826 * the given pipe inode to the given file.
827 *
828 */
Ingo Molnar3a326a22006-04-10 15:18:35 +0200829ssize_t
830generic_file_splice_write(struct pipe_inode_info *pipe, struct file *out,
Jens Axboecbb7e572006-04-11 14:57:50 +0200831 loff_t *ppos, size_t len, unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +0200832{
Jens Axboe4f6f0bd2006-04-02 23:04:46 +0200833 struct address_space *mapping = out->f_mapping;
Jens Axboe8c34e2d2006-10-17 19:43:22 +0200834 struct inode *inode = mapping->host;
Ingo Molnar3a326a22006-04-10 15:18:35 +0200835 ssize_t ret;
Jens Axboe8c34e2d2006-10-17 19:43:22 +0200836 int err;
837
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800838 err = should_remove_suid(out->f_path.dentry);
Jens Axboe8c34e2d2006-10-17 19:43:22 +0200839 if (unlikely(err)) {
840 mutex_lock(&inode->i_mutex);
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800841 err = __remove_suid(out->f_path.dentry, err);
Jens Axboe8c34e2d2006-10-17 19:43:22 +0200842 mutex_unlock(&inode->i_mutex);
843 if (err)
844 return err;
845 }
Ingo Molnar3a326a22006-04-10 15:18:35 +0200846
Jens Axboe00522fb2006-04-26 14:39:29 +0200847 ret = splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_file);
Jens Axboea4514eb2006-04-19 15:57:05 +0200848 if (ret > 0) {
Jens Axboea4514eb2006-04-19 15:57:05 +0200849 *ppos += ret;
Jens Axboe4f6f0bd2006-04-02 23:04:46 +0200850
Jens Axboea4514eb2006-04-19 15:57:05 +0200851 /*
852 * If file or inode is SYNC and we actually wrote some data,
853 * sync it.
854 */
855 if (unlikely((out->f_flags & O_SYNC) || IS_SYNC(inode))) {
Jens Axboea4514eb2006-04-19 15:57:05 +0200856 mutex_lock(&inode->i_mutex);
857 err = generic_osync_inode(inode, mapping,
858 OSYNC_METADATA|OSYNC_DATA);
859 mutex_unlock(&inode->i_mutex);
860
861 if (err)
862 ret = err;
863 }
Jens Axboe4f6f0bd2006-04-02 23:04:46 +0200864 }
865
866 return ret;
Jens Axboe5274f052006-03-30 15:15:30 +0200867}
868
Jens Axboe059a8f32006-04-02 23:06:05 +0200869EXPORT_SYMBOL(generic_file_splice_write);
870
Jens Axboe83f91352006-04-02 23:05:09 +0200871/**
872 * generic_splice_sendpage - splice data from a pipe to a socket
873 * @inode: pipe inode
874 * @out: socket to write to
875 * @len: number of bytes to splice
876 * @flags: splice modifier flags
877 *
878 * Will send @len bytes from the pipe to a network socket. No data copying
879 * is involved.
880 *
881 */
Ingo Molnar3a326a22006-04-10 15:18:35 +0200882ssize_t generic_splice_sendpage(struct pipe_inode_info *pipe, struct file *out,
Jens Axboecbb7e572006-04-11 14:57:50 +0200883 loff_t *ppos, size_t len, unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +0200884{
Jens Axboe00522fb2006-04-26 14:39:29 +0200885 return splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_sendpage);
Jens Axboe5274f052006-03-30 15:15:30 +0200886}
887
Jens Axboe059a8f32006-04-02 23:06:05 +0200888EXPORT_SYMBOL(generic_splice_sendpage);
Jeff Garzika0f06782006-03-30 23:06:13 -0500889
Jens Axboe83f91352006-04-02 23:05:09 +0200890/*
891 * Attempt to initiate a splice from pipe to file.
892 */
Ingo Molnar3a326a22006-04-10 15:18:35 +0200893static long do_splice_from(struct pipe_inode_info *pipe, struct file *out,
Jens Axboecbb7e572006-04-11 14:57:50 +0200894 loff_t *ppos, size_t len, unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +0200895{
Jens Axboe5274f052006-03-30 15:15:30 +0200896 int ret;
897
Jens Axboe49570e92006-04-11 13:56:09 +0200898 if (unlikely(!out->f_op || !out->f_op->splice_write))
Jens Axboe5274f052006-03-30 15:15:30 +0200899 return -EINVAL;
900
Jens Axboe49570e92006-04-11 13:56:09 +0200901 if (unlikely(!(out->f_mode & FMODE_WRITE)))
Jens Axboe5274f052006-03-30 15:15:30 +0200902 return -EBADF;
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 Axboecbb7e572006-04-11 14:57:50 +0200918 loff_t isize, left;
Jens Axboe5274f052006-03-30 15:15:30 +0200919 int ret;
920
Jens Axboe49570e92006-04-11 13:56:09 +0200921 if (unlikely(!in->f_op || !in->f_op->splice_read))
Jens Axboe5274f052006-03-30 15:15:30 +0200922 return -EINVAL;
923
Jens Axboe49570e92006-04-11 13:56:09 +0200924 if (unlikely(!(in->f_mode & FMODE_READ)))
Jens Axboe5274f052006-03-30 15:15:30 +0200925 return -EBADF;
926
Jens Axboecbb7e572006-04-11 14:57:50 +0200927 ret = rw_verify_area(READ, in, ppos, len);
Jens Axboe5274f052006-03-30 15:15:30 +0200928 if (unlikely(ret < 0))
929 return ret;
930
931 isize = i_size_read(in->f_mapping->host);
Jens Axboecbb7e572006-04-11 14:57:50 +0200932 if (unlikely(*ppos >= isize))
Jens Axboe5274f052006-03-30 15:15:30 +0200933 return 0;
934
Jens Axboecbb7e572006-04-11 14:57:50 +0200935 left = isize - *ppos;
Jens Axboe49570e92006-04-11 13:56:09 +0200936 if (unlikely(left < len))
Jens Axboe5274f052006-03-30 15:15:30 +0200937 len = left;
938
Jens Axboecbb7e572006-04-11 14:57:50 +0200939 return in->f_op->splice_read(in, ppos, pipe, len, flags);
Jens Axboe5274f052006-03-30 15:15:30 +0200940}
941
Jens Axboecbb7e572006-04-11 14:57:50 +0200942long do_splice_direct(struct file *in, loff_t *ppos, struct file *out,
943 size_t len, unsigned int flags)
Jens Axboeb92ce552006-04-11 13:52:07 +0200944{
945 struct pipe_inode_info *pipe;
946 long ret, bytes;
Jens Axboecbb7e572006-04-11 14:57:50 +0200947 loff_t out_off;
Jens Axboeb92ce552006-04-11 13:52:07 +0200948 umode_t i_mode;
949 int i;
950
951 /*
952 * We require the input being a regular file, as we don't want to
953 * randomly drop data for eg socket -> socket splicing. Use the
954 * piped splicing for that!
955 */
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800956 i_mode = in->f_path.dentry->d_inode->i_mode;
Jens Axboeb92ce552006-04-11 13:52:07 +0200957 if (unlikely(!S_ISREG(i_mode) && !S_ISBLK(i_mode)))
958 return -EINVAL;
959
960 /*
961 * neither in nor out is a pipe, setup an internal pipe attached to
962 * 'out' and transfer the wanted data from 'in' to 'out' through that
963 */
964 pipe = current->splice_pipe;
Jens Axboe49570e92006-04-11 13:56:09 +0200965 if (unlikely(!pipe)) {
Jens Axboeb92ce552006-04-11 13:52:07 +0200966 pipe = alloc_pipe_info(NULL);
967 if (!pipe)
968 return -ENOMEM;
969
970 /*
971 * We don't have an immediate reader, but we'll read the stuff
Jens Axboe00522fb2006-04-26 14:39:29 +0200972 * out of the pipe right after the splice_to_pipe(). So set
Jens Axboeb92ce552006-04-11 13:52:07 +0200973 * PIPE_READERS appropriately.
974 */
975 pipe->readers = 1;
976
977 current->splice_pipe = pipe;
978 }
979
980 /*
Ingo Molnar73d62d82006-04-11 13:57:21 +0200981 * Do the splice.
Jens Axboeb92ce552006-04-11 13:52:07 +0200982 */
983 ret = 0;
984 bytes = 0;
Jens Axboecbb7e572006-04-11 14:57:50 +0200985 out_off = 0;
Jens Axboeb92ce552006-04-11 13:52:07 +0200986
987 while (len) {
988 size_t read_len, max_read_len;
989
990 /*
991 * Do at most PIPE_BUFFERS pages worth of transfer:
992 */
993 max_read_len = min(len, (size_t)(PIPE_BUFFERS*PAGE_SIZE));
994
Jens Axboecbb7e572006-04-11 14:57:50 +0200995 ret = do_splice_to(in, ppos, pipe, max_read_len, flags);
Jens Axboeb92ce552006-04-11 13:52:07 +0200996 if (unlikely(ret < 0))
997 goto out_release;
998
999 read_len = ret;
1000
1001 /*
1002 * NOTE: nonblocking mode only applies to the input. We
1003 * must not do the output in nonblocking mode as then we
1004 * could get stuck data in the internal pipe:
1005 */
Jens Axboecbb7e572006-04-11 14:57:50 +02001006 ret = do_splice_from(pipe, out, &out_off, read_len,
Jens Axboeb92ce552006-04-11 13:52:07 +02001007 flags & ~SPLICE_F_NONBLOCK);
1008 if (unlikely(ret < 0))
1009 goto out_release;
1010
1011 bytes += ret;
1012 len -= ret;
1013
1014 /*
1015 * In nonblocking mode, if we got back a short read then
1016 * that was due to either an IO error or due to the
1017 * pagecache entry not being there. In the IO error case
1018 * the _next_ splice attempt will produce a clean IO error
1019 * return value (not a short read), so in both cases it's
1020 * correct to break out of the loop here:
1021 */
1022 if ((flags & SPLICE_F_NONBLOCK) && (read_len < max_read_len))
1023 break;
1024 }
1025
1026 pipe->nrbufs = pipe->curbuf = 0;
1027
1028 return bytes;
1029
1030out_release:
1031 /*
1032 * If we did an incomplete transfer we must release
1033 * the pipe buffers in question:
1034 */
1035 for (i = 0; i < PIPE_BUFFERS; i++) {
1036 struct pipe_buffer *buf = pipe->bufs + i;
1037
1038 if (buf->ops) {
1039 buf->ops->release(pipe, buf);
1040 buf->ops = NULL;
1041 }
1042 }
1043 pipe->nrbufs = pipe->curbuf = 0;
1044
1045 /*
1046 * If we transferred some data, return the number of bytes:
1047 */
1048 if (bytes > 0)
1049 return bytes;
1050
1051 return ret;
1052}
1053
1054EXPORT_SYMBOL(do_splice_direct);
1055
Jens Axboe83f91352006-04-02 23:05:09 +02001056/*
Jens Axboeddac0d32006-11-04 12:49:32 +01001057 * After the inode slimming patch, i_pipe/i_bdev/i_cdev share the same
1058 * location, so checking ->i_pipe is not enough to verify that this is a
1059 * pipe.
1060 */
1061static inline struct pipe_inode_info *pipe_info(struct inode *inode)
1062{
1063 if (S_ISFIFO(inode->i_mode))
1064 return inode->i_pipe;
1065
1066 return NULL;
1067}
1068
1069/*
Jens Axboe83f91352006-04-02 23:05:09 +02001070 * Determine where to splice to/from.
1071 */
Ingo Molnar529565d2006-04-10 15:18:58 +02001072static long do_splice(struct file *in, loff_t __user *off_in,
1073 struct file *out, loff_t __user *off_out,
1074 size_t len, unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +02001075{
Ingo Molnar3a326a22006-04-10 15:18:35 +02001076 struct pipe_inode_info *pipe;
Jens Axboecbb7e572006-04-11 14:57:50 +02001077 loff_t offset, *off;
Jens Axboea4514eb2006-04-19 15:57:05 +02001078 long ret;
Jens Axboe5274f052006-03-30 15:15:30 +02001079
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001080 pipe = pipe_info(in->f_path.dentry->d_inode);
Ingo Molnar529565d2006-04-10 15:18:58 +02001081 if (pipe) {
1082 if (off_in)
1083 return -ESPIPE;
Jens Axboeb92ce552006-04-11 13:52:07 +02001084 if (off_out) {
1085 if (out->f_op->llseek == no_llseek)
1086 return -EINVAL;
Jens Axboecbb7e572006-04-11 14:57:50 +02001087 if (copy_from_user(&offset, off_out, sizeof(loff_t)))
Jens Axboeb92ce552006-04-11 13:52:07 +02001088 return -EFAULT;
Jens Axboecbb7e572006-04-11 14:57:50 +02001089 off = &offset;
1090 } else
1091 off = &out->f_pos;
Ingo Molnar529565d2006-04-10 15:18:58 +02001092
Jens Axboea4514eb2006-04-19 15:57:05 +02001093 ret = do_splice_from(pipe, out, off, len, flags);
1094
1095 if (off_out && copy_to_user(off_out, off, sizeof(loff_t)))
1096 ret = -EFAULT;
1097
1098 return ret;
Ingo Molnar529565d2006-04-10 15:18:58 +02001099 }
Jens Axboe5274f052006-03-30 15:15:30 +02001100
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001101 pipe = pipe_info(out->f_path.dentry->d_inode);
Ingo Molnar529565d2006-04-10 15:18:58 +02001102 if (pipe) {
1103 if (off_out)
1104 return -ESPIPE;
Jens Axboeb92ce552006-04-11 13:52:07 +02001105 if (off_in) {
1106 if (in->f_op->llseek == no_llseek)
1107 return -EINVAL;
Jens Axboecbb7e572006-04-11 14:57:50 +02001108 if (copy_from_user(&offset, off_in, sizeof(loff_t)))
Jens Axboeb92ce552006-04-11 13:52:07 +02001109 return -EFAULT;
Jens Axboecbb7e572006-04-11 14:57:50 +02001110 off = &offset;
1111 } else
1112 off = &in->f_pos;
Ingo Molnar529565d2006-04-10 15:18:58 +02001113
Jens Axboea4514eb2006-04-19 15:57:05 +02001114 ret = do_splice_to(in, off, pipe, len, flags);
1115
1116 if (off_in && copy_to_user(off_in, off, sizeof(loff_t)))
1117 ret = -EFAULT;
1118
1119 return ret;
Ingo Molnar529565d2006-04-10 15:18:58 +02001120 }
Jens Axboe5274f052006-03-30 15:15:30 +02001121
1122 return -EINVAL;
1123}
1124
Jens Axboe912d35f2006-04-26 10:59:21 +02001125/*
1126 * Map an iov into an array of pages and offset/length tupples. With the
1127 * partial_page structure, we can map several non-contiguous ranges into
1128 * our ones pages[] map instead of splitting that operation into pieces.
1129 * Could easily be exported as a generic helper for other users, in which
1130 * case one would probably want to add a 'max_nr_pages' parameter as well.
1131 */
1132static int get_iovec_page_array(const struct iovec __user *iov,
1133 unsigned int nr_vecs, struct page **pages,
Jens Axboe7afa6fd2006-05-01 20:02:33 +02001134 struct partial_page *partial, int aligned)
Jens Axboe912d35f2006-04-26 10:59:21 +02001135{
1136 int buffers = 0, error = 0;
1137
1138 /*
1139 * It's ok to take the mmap_sem for reading, even
1140 * across a "get_user()".
1141 */
1142 down_read(&current->mm->mmap_sem);
1143
1144 while (nr_vecs) {
1145 unsigned long off, npages;
1146 void __user *base;
1147 size_t len;
1148 int i;
1149
1150 /*
1151 * Get user address base and length for this iovec.
1152 */
1153 error = get_user(base, &iov->iov_base);
1154 if (unlikely(error))
1155 break;
1156 error = get_user(len, &iov->iov_len);
1157 if (unlikely(error))
1158 break;
1159
1160 /*
1161 * Sanity check this iovec. 0 read succeeds.
1162 */
1163 if (unlikely(!len))
1164 break;
1165 error = -EFAULT;
1166 if (unlikely(!base))
1167 break;
1168
1169 /*
1170 * Get this base offset and number of pages, then map
1171 * in the user pages.
1172 */
1173 off = (unsigned long) base & ~PAGE_MASK;
Jens Axboe7afa6fd2006-05-01 20:02:33 +02001174
1175 /*
1176 * If asked for alignment, the offset must be zero and the
1177 * length a multiple of the PAGE_SIZE.
1178 */
1179 error = -EINVAL;
1180 if (aligned && (off || len & ~PAGE_MASK))
1181 break;
1182
Jens Axboe912d35f2006-04-26 10:59:21 +02001183 npages = (off + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1184 if (npages > PIPE_BUFFERS - buffers)
1185 npages = PIPE_BUFFERS - buffers;
1186
1187 error = get_user_pages(current, current->mm,
1188 (unsigned long) base, npages, 0, 0,
1189 &pages[buffers], NULL);
1190
1191 if (unlikely(error <= 0))
1192 break;
1193
1194 /*
1195 * Fill this contiguous range into the partial page map.
1196 */
1197 for (i = 0; i < error; i++) {
Jens Axboe75914892006-05-02 12:57:18 +02001198 const int plen = min_t(size_t, len, PAGE_SIZE - off);
Jens Axboe912d35f2006-04-26 10:59:21 +02001199
1200 partial[buffers].offset = off;
1201 partial[buffers].len = plen;
1202
1203 off = 0;
1204 len -= plen;
1205 buffers++;
1206 }
1207
1208 /*
1209 * We didn't complete this iov, stop here since it probably
1210 * means we have to move some of this into a pipe to
1211 * be able to continue.
1212 */
1213 if (len)
1214 break;
1215
1216 /*
1217 * Don't continue if we mapped fewer pages than we asked for,
1218 * or if we mapped the max number of pages that we have
1219 * room for.
1220 */
1221 if (error < npages || buffers == PIPE_BUFFERS)
1222 break;
1223
1224 nr_vecs--;
1225 iov++;
1226 }
1227
1228 up_read(&current->mm->mmap_sem);
1229
1230 if (buffers)
1231 return buffers;
1232
1233 return error;
1234}
1235
1236/*
1237 * vmsplice splices a user address range into a pipe. It can be thought of
1238 * as splice-from-memory, where the regular splice is splice-from-file (or
1239 * to file). In both cases the output is a pipe, naturally.
1240 *
1241 * Note that vmsplice only supports splicing _from_ user memory to a pipe,
1242 * not the other way around. Splicing from user memory is a simple operation
1243 * that can be supported without any funky alignment restrictions or nasty
1244 * vm tricks. We simply map in the user memory and fill them into a pipe.
1245 * The reverse isn't quite as easy, though. There are two possible solutions
1246 * for that:
1247 *
1248 * - memcpy() the data internally, at which point we might as well just
1249 * do a regular read() on the buffer anyway.
1250 * - Lots of nasty vm tricks, that are neither fast nor flexible (it
1251 * has restriction limitations on both ends of the pipe).
1252 *
1253 * Alas, it isn't here.
1254 *
1255 */
1256static long do_vmsplice(struct file *file, const struct iovec __user *iov,
1257 unsigned long nr_segs, unsigned int flags)
1258{
Jens Axboeddac0d32006-11-04 12:49:32 +01001259 struct pipe_inode_info *pipe;
Jens Axboe912d35f2006-04-26 10:59:21 +02001260 struct page *pages[PIPE_BUFFERS];
1261 struct partial_page partial[PIPE_BUFFERS];
1262 struct splice_pipe_desc spd = {
1263 .pages = pages,
1264 .partial = partial,
1265 .flags = flags,
1266 .ops = &user_page_pipe_buf_ops,
1267 };
1268
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001269 pipe = pipe_info(file->f_path.dentry->d_inode);
Jens Axboeddac0d32006-11-04 12:49:32 +01001270 if (!pipe)
Jens Axboe912d35f2006-04-26 10:59:21 +02001271 return -EBADF;
1272 if (unlikely(nr_segs > UIO_MAXIOV))
1273 return -EINVAL;
1274 else if (unlikely(!nr_segs))
1275 return 0;
1276
Jens Axboe7afa6fd2006-05-01 20:02:33 +02001277 spd.nr_pages = get_iovec_page_array(iov, nr_segs, pages, partial,
1278 flags & SPLICE_F_GIFT);
Jens Axboe912d35f2006-04-26 10:59:21 +02001279 if (spd.nr_pages <= 0)
1280 return spd.nr_pages;
1281
Jens Axboe00522fb2006-04-26 14:39:29 +02001282 return splice_to_pipe(pipe, &spd);
Jens Axboe912d35f2006-04-26 10:59:21 +02001283}
1284
1285asmlinkage long sys_vmsplice(int fd, const struct iovec __user *iov,
1286 unsigned long nr_segs, unsigned int flags)
1287{
1288 struct file *file;
1289 long error;
1290 int fput;
1291
1292 error = -EBADF;
1293 file = fget_light(fd, &fput);
1294 if (file) {
1295 if (file->f_mode & FMODE_WRITE)
1296 error = do_vmsplice(file, iov, nr_segs, flags);
1297
1298 fput_light(file, fput);
1299 }
1300
1301 return error;
1302}
1303
Ingo Molnar529565d2006-04-10 15:18:58 +02001304asmlinkage long sys_splice(int fd_in, loff_t __user *off_in,
1305 int fd_out, loff_t __user *off_out,
1306 size_t len, unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +02001307{
1308 long error;
1309 struct file *in, *out;
1310 int fput_in, fput_out;
1311
1312 if (unlikely(!len))
1313 return 0;
1314
1315 error = -EBADF;
Ingo Molnar529565d2006-04-10 15:18:58 +02001316 in = fget_light(fd_in, &fput_in);
Jens Axboe5274f052006-03-30 15:15:30 +02001317 if (in) {
1318 if (in->f_mode & FMODE_READ) {
Ingo Molnar529565d2006-04-10 15:18:58 +02001319 out = fget_light(fd_out, &fput_out);
Jens Axboe5274f052006-03-30 15:15:30 +02001320 if (out) {
1321 if (out->f_mode & FMODE_WRITE)
Ingo Molnar529565d2006-04-10 15:18:58 +02001322 error = do_splice(in, off_in,
1323 out, off_out,
1324 len, flags);
Jens Axboe5274f052006-03-30 15:15:30 +02001325 fput_light(out, fput_out);
1326 }
1327 }
1328
1329 fput_light(in, fput_in);
1330 }
1331
1332 return error;
1333}
Jens Axboe70524492006-04-11 15:51:17 +02001334
1335/*
Jens Axboeaadd06e2006-07-10 11:00:01 +02001336 * Make sure there's data to read. Wait for input if we can, otherwise
1337 * return an appropriate error.
1338 */
1339static int link_ipipe_prep(struct pipe_inode_info *pipe, unsigned int flags)
1340{
1341 int ret;
1342
1343 /*
1344 * Check ->nrbufs without the inode lock first. This function
1345 * is speculative anyways, so missing one is ok.
1346 */
1347 if (pipe->nrbufs)
1348 return 0;
1349
1350 ret = 0;
1351 mutex_lock(&pipe->inode->i_mutex);
1352
1353 while (!pipe->nrbufs) {
1354 if (signal_pending(current)) {
1355 ret = -ERESTARTSYS;
1356 break;
1357 }
1358 if (!pipe->writers)
1359 break;
1360 if (!pipe->waiting_writers) {
1361 if (flags & SPLICE_F_NONBLOCK) {
1362 ret = -EAGAIN;
1363 break;
1364 }
1365 }
1366 pipe_wait(pipe);
1367 }
1368
1369 mutex_unlock(&pipe->inode->i_mutex);
1370 return ret;
1371}
1372
1373/*
1374 * Make sure there's writeable room. Wait for room if we can, otherwise
1375 * return an appropriate error.
1376 */
1377static int link_opipe_prep(struct pipe_inode_info *pipe, unsigned int flags)
1378{
1379 int ret;
1380
1381 /*
1382 * Check ->nrbufs without the inode lock first. This function
1383 * is speculative anyways, so missing one is ok.
1384 */
1385 if (pipe->nrbufs < PIPE_BUFFERS)
1386 return 0;
1387
1388 ret = 0;
1389 mutex_lock(&pipe->inode->i_mutex);
1390
1391 while (pipe->nrbufs >= PIPE_BUFFERS) {
1392 if (!pipe->readers) {
1393 send_sig(SIGPIPE, current, 0);
1394 ret = -EPIPE;
1395 break;
1396 }
1397 if (flags & SPLICE_F_NONBLOCK) {
1398 ret = -EAGAIN;
1399 break;
1400 }
1401 if (signal_pending(current)) {
1402 ret = -ERESTARTSYS;
1403 break;
1404 }
1405 pipe->waiting_writers++;
1406 pipe_wait(pipe);
1407 pipe->waiting_writers--;
1408 }
1409
1410 mutex_unlock(&pipe->inode->i_mutex);
1411 return ret;
1412}
1413
1414/*
Jens Axboe70524492006-04-11 15:51:17 +02001415 * Link contents of ipipe to opipe.
1416 */
1417static int link_pipe(struct pipe_inode_info *ipipe,
1418 struct pipe_inode_info *opipe,
1419 size_t len, unsigned int flags)
1420{
1421 struct pipe_buffer *ibuf, *obuf;
Jens Axboeaadd06e2006-07-10 11:00:01 +02001422 int ret = 0, i = 0, nbuf;
Jens Axboe70524492006-04-11 15:51:17 +02001423
1424 /*
1425 * Potential ABBA deadlock, work around it by ordering lock
1426 * grabbing by inode address. Otherwise two different processes
1427 * could deadlock (one doing tee from A -> B, the other from B -> A).
1428 */
Mark Fasheh62752ee2006-10-17 10:31:38 +02001429 inode_double_lock(ipipe->inode, opipe->inode);
Jens Axboe70524492006-04-11 15:51:17 +02001430
Jens Axboeaadd06e2006-07-10 11:00:01 +02001431 do {
Jens Axboe70524492006-04-11 15:51:17 +02001432 if (!opipe->readers) {
1433 send_sig(SIGPIPE, current, 0);
1434 if (!ret)
1435 ret = -EPIPE;
1436 break;
1437 }
Jens Axboe70524492006-04-11 15:51:17 +02001438
1439 /*
Jens Axboeaadd06e2006-07-10 11:00:01 +02001440 * If we have iterated all input buffers or ran out of
1441 * output room, break.
Jens Axboe70524492006-04-11 15:51:17 +02001442 */
Jens Axboeaadd06e2006-07-10 11:00:01 +02001443 if (i >= ipipe->nrbufs || opipe->nrbufs >= PIPE_BUFFERS)
Jens Axboe70524492006-04-11 15:51:17 +02001444 break;
Jens Axboeaadd06e2006-07-10 11:00:01 +02001445
1446 ibuf = ipipe->bufs + ((ipipe->curbuf + i) & (PIPE_BUFFERS - 1));
1447 nbuf = (opipe->curbuf + opipe->nrbufs) & (PIPE_BUFFERS - 1);
1448
Jens Axboe2a272502006-04-19 15:56:40 +02001449 /*
Jens Axboeaadd06e2006-07-10 11:00:01 +02001450 * Get a reference to this pipe buffer,
1451 * so we can copy the contents over.
Jens Axboe2a272502006-04-19 15:56:40 +02001452 */
Jens Axboeaadd06e2006-07-10 11:00:01 +02001453 ibuf->ops->get(ipipe, ibuf);
Jens Axboe70524492006-04-11 15:51:17 +02001454
Jens Axboeaadd06e2006-07-10 11:00:01 +02001455 obuf = opipe->bufs + nbuf;
1456 *obuf = *ibuf;
Jens Axboe70524492006-04-11 15:51:17 +02001457
Jens Axboeaadd06e2006-07-10 11:00:01 +02001458 /*
1459 * Don't inherit the gift flag, we need to
1460 * prevent multiple steals of this page.
1461 */
1462 obuf->flags &= ~PIPE_BUF_FLAG_GIFT;
1463
1464 if (obuf->len > len)
1465 obuf->len = len;
1466
1467 opipe->nrbufs++;
1468 ret += obuf->len;
1469 len -= obuf->len;
1470 i++;
1471 } while (len);
Jens Axboe70524492006-04-11 15:51:17 +02001472
Mark Fasheh62752ee2006-10-17 10:31:38 +02001473 inode_double_unlock(ipipe->inode, opipe->inode);
Jens Axboe70524492006-04-11 15:51:17 +02001474
Jens Axboeaadd06e2006-07-10 11:00:01 +02001475 /*
1476 * If we put data in the output pipe, wakeup any potential readers.
1477 */
1478 if (ret > 0) {
Jens Axboe70524492006-04-11 15:51:17 +02001479 smp_mb();
1480 if (waitqueue_active(&opipe->wait))
1481 wake_up_interruptible(&opipe->wait);
1482 kill_fasync(&opipe->fasync_readers, SIGIO, POLL_IN);
1483 }
1484
1485 return ret;
1486}
1487
1488/*
1489 * This is a tee(1) implementation that works on pipes. It doesn't copy
1490 * any data, it simply references the 'in' pages on the 'out' pipe.
1491 * The 'flags' used are the SPLICE_F_* variants, currently the only
1492 * applicable one is SPLICE_F_NONBLOCK.
1493 */
1494static long do_tee(struct file *in, struct file *out, size_t len,
1495 unsigned int flags)
1496{
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001497 struct pipe_inode_info *ipipe = pipe_info(in->f_path.dentry->d_inode);
1498 struct pipe_inode_info *opipe = pipe_info(out->f_path.dentry->d_inode);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001499 int ret = -EINVAL;
Jens Axboe70524492006-04-11 15:51:17 +02001500
1501 /*
Jens Axboeaadd06e2006-07-10 11:00:01 +02001502 * Duplicate the contents of ipipe to opipe without actually
1503 * copying the data.
Jens Axboe70524492006-04-11 15:51:17 +02001504 */
Jens Axboeaadd06e2006-07-10 11:00:01 +02001505 if (ipipe && opipe && ipipe != opipe) {
1506 /*
1507 * Keep going, unless we encounter an error. The ipipe/opipe
1508 * ordering doesn't really matter.
1509 */
1510 ret = link_ipipe_prep(ipipe, flags);
1511 if (!ret) {
1512 ret = link_opipe_prep(opipe, flags);
1513 if (!ret) {
1514 ret = link_pipe(ipipe, opipe, len, flags);
1515 if (!ret && (flags & SPLICE_F_NONBLOCK))
1516 ret = -EAGAIN;
1517 }
1518 }
1519 }
Jens Axboe70524492006-04-11 15:51:17 +02001520
Jens Axboeaadd06e2006-07-10 11:00:01 +02001521 return ret;
Jens Axboe70524492006-04-11 15:51:17 +02001522}
1523
1524asmlinkage long sys_tee(int fdin, int fdout, size_t len, unsigned int flags)
1525{
1526 struct file *in;
1527 int error, fput_in;
1528
1529 if (unlikely(!len))
1530 return 0;
1531
1532 error = -EBADF;
1533 in = fget_light(fdin, &fput_in);
1534 if (in) {
1535 if (in->f_mode & FMODE_READ) {
1536 int fput_out;
1537 struct file *out = fget_light(fdout, &fput_out);
1538
1539 if (out) {
1540 if (out->f_mode & FMODE_WRITE)
1541 error = do_tee(in, out, len, flags);
1542 fput_light(out, fput_out);
1543 }
1544 }
1545 fput_light(in, fput_in);
1546 }
1547
1548 return error;
1549}