blob: bbd0aeb3f68e1723cc11c4be939137ebabb61dfd [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 */
45 struct pipe_buf_operations *ops;/* ops associated with output pipe */
46};
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
Jens Axboe5274f052006-03-30 15:15:30 +0200142static struct pipe_buf_operations page_cache_pipe_buf_ops = {
143 .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
162static struct pipe_buf_operations user_page_pipe_buf_ops = {
163 .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 Axboe5274f052006-03-30 15:15:30 +0200579 /*
Jens Axboe0568b402006-05-01 19:50:48 +0200580 * Reuse buf page, if SPLICE_F_MOVE is set and we are doing a full
581 * page.
Jens Axboe5274f052006-03-30 15:15:30 +0200582 */
Jens Axboe0568b402006-05-01 19:50:48 +0200583 if ((sd->flags & SPLICE_F_MOVE) && this_len == PAGE_CACHE_SIZE) {
Jens Axboe83f91352006-04-02 23:05:09 +0200584 /*
Jens Axboe14328732006-05-03 10:35:26 +0200585 * If steal succeeds, buf->page is now pruned from the
586 * pagecache and we can reuse it. The page will also be
587 * locked on successful return.
Jens Axboe83f91352006-04-02 23:05:09 +0200588 */
Jens Axboe76ad4d12006-05-03 10:41:33 +0200589 if (buf->ops->steal(pipe, buf))
Jens Axboe5abc97a2006-03-30 15:16:46 +0200590 goto find_page;
Jens Axboe5274f052006-03-30 15:15:30 +0200591
Jens Axboe5abc97a2006-03-30 15:16:46 +0200592 page = buf->page;
Nick Piggin2ae88142006-10-28 10:38:23 -0700593 if (add_to_page_cache(page, mapping, index, GFP_KERNEL)) {
Jens Axboe46e678c2006-04-30 16:36:32 +0200594 unlock_page(page);
Jens Axboe5abc97a2006-03-30 15:16:46 +0200595 goto find_page;
Jens Axboe46e678c2006-04-30 16:36:32 +0200596 }
Jens Axboe14328732006-05-03 10:35:26 +0200597
598 page_cache_get(page);
599
600 if (!(buf->flags & PIPE_BUF_FLAG_LRU))
601 lru_cache_add(page);
Jens Axboe5abc97a2006-03-30 15:16:46 +0200602 } else {
603find_page:
Jens Axboe9e0267c2006-04-19 15:57:31 +0200604 page = find_lock_page(mapping, index);
605 if (!page) {
606 ret = -ENOMEM;
607 page = page_cache_alloc_cold(mapping);
608 if (unlikely(!page))
Jens Axboee6e80f22006-10-11 10:03:09 +0200609 goto out_ret;
Jens Axboe9e0267c2006-04-19 15:57:31 +0200610
611 /*
612 * This will also lock the page
613 */
614 ret = add_to_page_cache_lru(page, mapping, index,
Nick Piggin2ae88142006-10-28 10:38:23 -0700615 GFP_KERNEL);
Jens Axboe9e0267c2006-04-19 15:57:31 +0200616 if (unlikely(ret))
617 goto out;
618 }
Jens Axboe5274f052006-03-30 15:15:30 +0200619
Jens Axboe5abc97a2006-03-30 15:16:46 +0200620 /*
Jens Axboe9e0267c2006-04-19 15:57:31 +0200621 * We get here with the page locked. If the page is also
622 * uptodate, we don't need to do more. If it isn't, we
623 * may need to bring it in if we are not going to overwrite
624 * the full page.
Jens Axboe5abc97a2006-03-30 15:16:46 +0200625 */
626 if (!PageUptodate(page)) {
Jens Axboe016b6612006-04-25 15:42:00 +0200627 if (this_len < PAGE_CACHE_SIZE) {
Jens Axboe5abc97a2006-03-30 15:16:46 +0200628 ret = mapping->a_ops->readpage(file, page);
629 if (unlikely(ret))
630 goto out;
631
632 lock_page(page);
633
634 if (!PageUptodate(page)) {
635 /*
Ingo Molnar73d62d82006-04-11 13:57:21 +0200636 * Page got invalidated, repeat.
Jens Axboe5abc97a2006-03-30 15:16:46 +0200637 */
638 if (!page->mapping) {
639 unlock_page(page);
640 page_cache_release(page);
641 goto find_page;
642 }
643 ret = -EIO;
644 goto out;
Jens Axboe5274f052006-03-30 15:15:30 +0200645 }
Jens Axboe9e0267c2006-04-19 15:57:31 +0200646 } else
Jens Axboe5abc97a2006-03-30 15:16:46 +0200647 SetPageUptodate(page);
Jens Axboe5274f052006-03-30 15:15:30 +0200648 }
649 }
650
Jens Axboe016b6612006-04-25 15:42:00 +0200651 ret = mapping->a_ops->prepare_write(file, page, offset, offset+this_len);
Jens Axboebfc4ee32006-05-03 10:35:10 +0200652 if (unlikely(ret)) {
653 loff_t isize = i_size_read(mapping->host);
654
655 if (ret != AOP_TRUNCATED_PAGE)
656 unlock_page(page);
Jens Axboe4f6f0bd2006-04-02 23:04:46 +0200657 page_cache_release(page);
Jens Axboebfc4ee32006-05-03 10:35:10 +0200658 if (ret == AOP_TRUNCATED_PAGE)
659 goto find_page;
660
661 /*
662 * prepare_write() may have instantiated a few blocks
663 * outside i_size. Trim these off again.
664 */
665 if (sd->pos + this_len > isize)
666 vmtruncate(mapping->host, isize);
667
Jens Axboee6e80f22006-10-11 10:03:09 +0200668 goto out_ret;
Jens Axboebfc4ee32006-05-03 10:35:10 +0200669 }
Jens Axboe5274f052006-03-30 15:15:30 +0200670
Jens Axboe0568b402006-05-01 19:50:48 +0200671 if (buf->page != page) {
Jens Axboef84d7512006-05-01 19:59:03 +0200672 /*
673 * Careful, ->map() uses KM_USER0!
674 */
Jens Axboe76ad4d12006-05-03 10:41:33 +0200675 char *src = buf->ops->map(pipe, buf, 1);
Jens Axboef84d7512006-05-01 19:59:03 +0200676 char *dst = kmap_atomic(page, KM_USER1);
Jens Axboe5abc97a2006-03-30 15:16:46 +0200677
Jens Axboe016b6612006-04-25 15:42:00 +0200678 memcpy(dst + offset, src + buf->offset, this_len);
Jens Axboe5abc97a2006-03-30 15:16:46 +0200679 flush_dcache_page(page);
Jens Axboef84d7512006-05-01 19:59:03 +0200680 kunmap_atomic(dst, KM_USER1);
Jens Axboe76ad4d12006-05-03 10:41:33 +0200681 buf->ops->unmap(pipe, buf, src);
Jens Axboe5abc97a2006-03-30 15:16:46 +0200682 }
Jens Axboe5274f052006-03-30 15:15:30 +0200683
Jens Axboe016b6612006-04-25 15:42:00 +0200684 ret = mapping->a_ops->commit_write(file, page, offset, offset+this_len);
Jens Axboe0568b402006-05-01 19:50:48 +0200685 if (!ret) {
686 /*
687 * Return the number of bytes written and mark page as
688 * accessed, we are now done!
689 */
690 ret = this_len;
691 mark_page_accessed(page);
692 balance_dirty_pages_ratelimited(mapping);
693 } else if (ret == AOP_TRUNCATED_PAGE) {
Jens Axboe4f6f0bd2006-04-02 23:04:46 +0200694 page_cache_release(page);
695 goto find_page;
Jens Axboe0568b402006-05-01 19:50:48 +0200696 }
Jens Axboe5274f052006-03-30 15:15:30 +0200697out:
Jens Axboe0568b402006-05-01 19:50:48 +0200698 page_cache_release(page);
Jens Axboe9e0267c2006-04-19 15:57:31 +0200699 unlock_page(page);
Jens Axboee6e80f22006-10-11 10:03:09 +0200700out_ret:
Jens Axboe5274f052006-03-30 15:15:30 +0200701 return ret;
702}
703
Jens Axboe83f91352006-04-02 23:05:09 +0200704/*
705 * Pipe input worker. Most of this logic works like a regular pipe, the
706 * key here is the 'actor' worker passed in that actually moves the data
707 * to the wanted destination. See pipe_to_file/pipe_to_sendpage above.
708 */
Mark Fasheh6da61802006-10-17 18:43:07 +0200709static ssize_t __splice_from_pipe(struct pipe_inode_info *pipe,
710 struct file *out, loff_t *ppos, size_t len,
711 unsigned int flags, splice_actor *actor)
Jens Axboe5274f052006-03-30 15:15:30 +0200712{
Jens Axboe5274f052006-03-30 15:15:30 +0200713 int ret, do_wakeup, err;
714 struct splice_desc sd;
715
716 ret = 0;
717 do_wakeup = 0;
718
719 sd.total_len = len;
720 sd.flags = flags;
721 sd.file = out;
Jens Axboecbb7e572006-04-11 14:57:50 +0200722 sd.pos = *ppos;
Jens Axboe5274f052006-03-30 15:15:30 +0200723
Jens Axboe5274f052006-03-30 15:15:30 +0200724 for (;;) {
Jens Axboe6f767b02006-04-11 13:53:56 +0200725 if (pipe->nrbufs) {
726 struct pipe_buffer *buf = pipe->bufs + pipe->curbuf;
Jens Axboe5274f052006-03-30 15:15:30 +0200727 struct pipe_buf_operations *ops = buf->ops;
728
729 sd.len = buf->len;
730 if (sd.len > sd.total_len)
731 sd.len = sd.total_len;
732
Ingo Molnar3a326a22006-04-10 15:18:35 +0200733 err = actor(pipe, buf, &sd);
Jens Axboe016b6612006-04-25 15:42:00 +0200734 if (err <= 0) {
Jens Axboe5274f052006-03-30 15:15:30 +0200735 if (!ret && err != -ENODATA)
736 ret = err;
737
738 break;
739 }
740
Jens Axboe016b6612006-04-25 15:42:00 +0200741 ret += err;
742 buf->offset += err;
743 buf->len -= err;
744
745 sd.len -= err;
746 sd.pos += err;
747 sd.total_len -= err;
748 if (sd.len)
749 continue;
Ingo Molnar73d62d82006-04-11 13:57:21 +0200750
Jens Axboe5274f052006-03-30 15:15:30 +0200751 if (!buf->len) {
752 buf->ops = NULL;
Ingo Molnar3a326a22006-04-10 15:18:35 +0200753 ops->release(pipe, buf);
Jens Axboe6f767b02006-04-11 13:53:56 +0200754 pipe->curbuf = (pipe->curbuf + 1) & (PIPE_BUFFERS - 1);
755 pipe->nrbufs--;
756 if (pipe->inode)
757 do_wakeup = 1;
Jens Axboe5274f052006-03-30 15:15:30 +0200758 }
759
Jens Axboe5274f052006-03-30 15:15:30 +0200760 if (!sd.total_len)
761 break;
762 }
763
Jens Axboe6f767b02006-04-11 13:53:56 +0200764 if (pipe->nrbufs)
Jens Axboe5274f052006-03-30 15:15:30 +0200765 continue;
Ingo Molnar3a326a22006-04-10 15:18:35 +0200766 if (!pipe->writers)
Jens Axboe5274f052006-03-30 15:15:30 +0200767 break;
Ingo Molnar3a326a22006-04-10 15:18:35 +0200768 if (!pipe->waiting_writers) {
Jens Axboe5274f052006-03-30 15:15:30 +0200769 if (ret)
770 break;
771 }
772
Linus Torvalds29e35092006-04-02 12:46:35 -0700773 if (flags & SPLICE_F_NONBLOCK) {
774 if (!ret)
775 ret = -EAGAIN;
776 break;
777 }
778
Jens Axboe5274f052006-03-30 15:15:30 +0200779 if (signal_pending(current)) {
780 if (!ret)
781 ret = -ERESTARTSYS;
782 break;
783 }
784
785 if (do_wakeup) {
Jens Axboec0bd1f62006-04-10 09:03:32 +0200786 smp_mb();
Ingo Molnar3a326a22006-04-10 15:18:35 +0200787 if (waitqueue_active(&pipe->wait))
788 wake_up_interruptible_sync(&pipe->wait);
789 kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
Jens Axboe5274f052006-03-30 15:15:30 +0200790 do_wakeup = 0;
791 }
792
Ingo Molnar3a326a22006-04-10 15:18:35 +0200793 pipe_wait(pipe);
Jens Axboe5274f052006-03-30 15:15:30 +0200794 }
795
Jens Axboe5274f052006-03-30 15:15:30 +0200796 if (do_wakeup) {
Jens Axboec0bd1f62006-04-10 09:03:32 +0200797 smp_mb();
Ingo Molnar3a326a22006-04-10 15:18:35 +0200798 if (waitqueue_active(&pipe->wait))
799 wake_up_interruptible(&pipe->wait);
800 kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
Jens Axboe5274f052006-03-30 15:15:30 +0200801 }
802
Jens Axboe5274f052006-03-30 15:15:30 +0200803 return ret;
Jens Axboe5274f052006-03-30 15:15:30 +0200804}
805
Mark Fasheh6da61802006-10-17 18:43:07 +0200806ssize_t splice_from_pipe(struct pipe_inode_info *pipe, struct file *out,
807 loff_t *ppos, size_t len, unsigned int flags,
808 splice_actor *actor)
809{
810 ssize_t ret;
811 struct inode *inode = out->f_mapping->host;
812
813 /*
814 * The actor worker might be calling ->prepare_write and
815 * ->commit_write. Most of the time, these expect i_mutex to
816 * be held. Since this may result in an ABBA deadlock with
817 * pipe->inode, we have to order lock acquiry here.
818 */
819 inode_double_lock(inode, pipe->inode);
820 ret = __splice_from_pipe(pipe, out, ppos, len, flags, actor);
821 inode_double_unlock(inode, pipe->inode);
822
823 return ret;
824}
825
826/**
827 * generic_file_splice_write_nolock - generic_file_splice_write without mutexes
828 * @pipe: pipe info
829 * @out: file to write to
830 * @len: number of bytes to splice
831 * @flags: splice modifier flags
832 *
833 * Will either move or copy pages (determined by @flags options) from
834 * the given pipe inode to the given file. The caller is responsible
835 * for acquiring i_mutex on both inodes.
836 *
837 */
838ssize_t
839generic_file_splice_write_nolock(struct pipe_inode_info *pipe, struct file *out,
840 loff_t *ppos, size_t len, unsigned int flags)
841{
842 struct address_space *mapping = out->f_mapping;
843 struct inode *inode = mapping->host;
844 ssize_t ret;
845 int err;
846
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800847 err = remove_suid(out->f_path.dentry);
Jens Axboe8c34e2d2006-10-17 19:43:22 +0200848 if (unlikely(err))
849 return err;
850
Mark Fasheh6da61802006-10-17 18:43:07 +0200851 ret = __splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_file);
852 if (ret > 0) {
853 *ppos += ret;
854
855 /*
856 * If file or inode is SYNC and we actually wrote some data,
857 * sync it.
858 */
859 if (unlikely((out->f_flags & O_SYNC) || IS_SYNC(inode))) {
860 err = generic_osync_inode(inode, mapping,
861 OSYNC_METADATA|OSYNC_DATA);
862
863 if (err)
864 ret = err;
865 }
866 }
867
868 return ret;
869}
870
871EXPORT_SYMBOL(generic_file_splice_write_nolock);
872
Jens Axboe83f91352006-04-02 23:05:09 +0200873/**
874 * generic_file_splice_write - splice data from a pipe to a file
Ingo Molnar3a326a22006-04-10 15:18:35 +0200875 * @pipe: pipe info
Jens Axboe83f91352006-04-02 23:05:09 +0200876 * @out: file to write to
877 * @len: number of bytes to splice
878 * @flags: splice modifier flags
879 *
880 * Will either move or copy pages (determined by @flags options) from
881 * the given pipe inode to the given file.
882 *
883 */
Ingo Molnar3a326a22006-04-10 15:18:35 +0200884ssize_t
885generic_file_splice_write(struct pipe_inode_info *pipe, struct file *out,
Jens Axboecbb7e572006-04-11 14:57:50 +0200886 loff_t *ppos, size_t len, unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +0200887{
Jens Axboe4f6f0bd2006-04-02 23:04:46 +0200888 struct address_space *mapping = out->f_mapping;
Jens Axboe8c34e2d2006-10-17 19:43:22 +0200889 struct inode *inode = mapping->host;
Ingo Molnar3a326a22006-04-10 15:18:35 +0200890 ssize_t ret;
Jens Axboe8c34e2d2006-10-17 19:43:22 +0200891 int err;
892
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800893 err = should_remove_suid(out->f_path.dentry);
Jens Axboe8c34e2d2006-10-17 19:43:22 +0200894 if (unlikely(err)) {
895 mutex_lock(&inode->i_mutex);
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800896 err = __remove_suid(out->f_path.dentry, err);
Jens Axboe8c34e2d2006-10-17 19:43:22 +0200897 mutex_unlock(&inode->i_mutex);
898 if (err)
899 return err;
900 }
Ingo Molnar3a326a22006-04-10 15:18:35 +0200901
Jens Axboe00522fb2006-04-26 14:39:29 +0200902 ret = splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_file);
Jens Axboea4514eb2006-04-19 15:57:05 +0200903 if (ret > 0) {
Jens Axboea4514eb2006-04-19 15:57:05 +0200904 *ppos += ret;
Jens Axboe4f6f0bd2006-04-02 23:04:46 +0200905
Jens Axboea4514eb2006-04-19 15:57:05 +0200906 /*
907 * If file or inode is SYNC and we actually wrote some data,
908 * sync it.
909 */
910 if (unlikely((out->f_flags & O_SYNC) || IS_SYNC(inode))) {
Jens Axboea4514eb2006-04-19 15:57:05 +0200911 mutex_lock(&inode->i_mutex);
912 err = generic_osync_inode(inode, mapping,
913 OSYNC_METADATA|OSYNC_DATA);
914 mutex_unlock(&inode->i_mutex);
915
916 if (err)
917 ret = err;
918 }
Jens Axboe4f6f0bd2006-04-02 23:04:46 +0200919 }
920
921 return ret;
Jens Axboe5274f052006-03-30 15:15:30 +0200922}
923
Jens Axboe059a8f32006-04-02 23:06:05 +0200924EXPORT_SYMBOL(generic_file_splice_write);
925
Jens Axboe83f91352006-04-02 23:05:09 +0200926/**
927 * generic_splice_sendpage - splice data from a pipe to a socket
928 * @inode: pipe inode
929 * @out: socket to write to
930 * @len: number of bytes to splice
931 * @flags: splice modifier flags
932 *
933 * Will send @len bytes from the pipe to a network socket. No data copying
934 * is involved.
935 *
936 */
Ingo Molnar3a326a22006-04-10 15:18:35 +0200937ssize_t generic_splice_sendpage(struct pipe_inode_info *pipe, struct file *out,
Jens Axboecbb7e572006-04-11 14:57:50 +0200938 loff_t *ppos, size_t len, unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +0200939{
Jens Axboe00522fb2006-04-26 14:39:29 +0200940 return splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_sendpage);
Jens Axboe5274f052006-03-30 15:15:30 +0200941}
942
Jens Axboe059a8f32006-04-02 23:06:05 +0200943EXPORT_SYMBOL(generic_splice_sendpage);
Jeff Garzika0f06782006-03-30 23:06:13 -0500944
Jens Axboe83f91352006-04-02 23:05:09 +0200945/*
946 * Attempt to initiate a splice from pipe to file.
947 */
Ingo Molnar3a326a22006-04-10 15:18:35 +0200948static long do_splice_from(struct pipe_inode_info *pipe, struct file *out,
Jens Axboecbb7e572006-04-11 14:57:50 +0200949 loff_t *ppos, size_t len, unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +0200950{
Jens Axboe5274f052006-03-30 15:15:30 +0200951 int ret;
952
Jens Axboe49570e92006-04-11 13:56:09 +0200953 if (unlikely(!out->f_op || !out->f_op->splice_write))
Jens Axboe5274f052006-03-30 15:15:30 +0200954 return -EINVAL;
955
Jens Axboe49570e92006-04-11 13:56:09 +0200956 if (unlikely(!(out->f_mode & FMODE_WRITE)))
Jens Axboe5274f052006-03-30 15:15:30 +0200957 return -EBADF;
958
Jens Axboecbb7e572006-04-11 14:57:50 +0200959 ret = rw_verify_area(WRITE, out, ppos, len);
Jens Axboe5274f052006-03-30 15:15:30 +0200960 if (unlikely(ret < 0))
961 return ret;
962
Jens Axboecbb7e572006-04-11 14:57:50 +0200963 return out->f_op->splice_write(pipe, out, ppos, len, flags);
Jens Axboe5274f052006-03-30 15:15:30 +0200964}
965
Jens Axboe83f91352006-04-02 23:05:09 +0200966/*
967 * Attempt to initiate a splice from a file to a pipe.
968 */
Jens Axboecbb7e572006-04-11 14:57:50 +0200969static long do_splice_to(struct file *in, loff_t *ppos,
970 struct pipe_inode_info *pipe, size_t len,
971 unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +0200972{
Jens Axboecbb7e572006-04-11 14:57:50 +0200973 loff_t isize, left;
Jens Axboe5274f052006-03-30 15:15:30 +0200974 int ret;
975
Jens Axboe49570e92006-04-11 13:56:09 +0200976 if (unlikely(!in->f_op || !in->f_op->splice_read))
Jens Axboe5274f052006-03-30 15:15:30 +0200977 return -EINVAL;
978
Jens Axboe49570e92006-04-11 13:56:09 +0200979 if (unlikely(!(in->f_mode & FMODE_READ)))
Jens Axboe5274f052006-03-30 15:15:30 +0200980 return -EBADF;
981
Jens Axboecbb7e572006-04-11 14:57:50 +0200982 ret = rw_verify_area(READ, in, ppos, len);
Jens Axboe5274f052006-03-30 15:15:30 +0200983 if (unlikely(ret < 0))
984 return ret;
985
986 isize = i_size_read(in->f_mapping->host);
Jens Axboecbb7e572006-04-11 14:57:50 +0200987 if (unlikely(*ppos >= isize))
Jens Axboe5274f052006-03-30 15:15:30 +0200988 return 0;
989
Jens Axboecbb7e572006-04-11 14:57:50 +0200990 left = isize - *ppos;
Jens Axboe49570e92006-04-11 13:56:09 +0200991 if (unlikely(left < len))
Jens Axboe5274f052006-03-30 15:15:30 +0200992 len = left;
993
Jens Axboecbb7e572006-04-11 14:57:50 +0200994 return in->f_op->splice_read(in, ppos, pipe, len, flags);
Jens Axboe5274f052006-03-30 15:15:30 +0200995}
996
Jens Axboecbb7e572006-04-11 14:57:50 +0200997long do_splice_direct(struct file *in, loff_t *ppos, struct file *out,
998 size_t len, unsigned int flags)
Jens Axboeb92ce552006-04-11 13:52:07 +0200999{
1000 struct pipe_inode_info *pipe;
1001 long ret, bytes;
Jens Axboecbb7e572006-04-11 14:57:50 +02001002 loff_t out_off;
Jens Axboeb92ce552006-04-11 13:52:07 +02001003 umode_t i_mode;
1004 int i;
1005
1006 /*
1007 * We require the input being a regular file, as we don't want to
1008 * randomly drop data for eg socket -> socket splicing. Use the
1009 * piped splicing for that!
1010 */
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001011 i_mode = in->f_path.dentry->d_inode->i_mode;
Jens Axboeb92ce552006-04-11 13:52:07 +02001012 if (unlikely(!S_ISREG(i_mode) && !S_ISBLK(i_mode)))
1013 return -EINVAL;
1014
1015 /*
1016 * neither in nor out is a pipe, setup an internal pipe attached to
1017 * 'out' and transfer the wanted data from 'in' to 'out' through that
1018 */
1019 pipe = current->splice_pipe;
Jens Axboe49570e92006-04-11 13:56:09 +02001020 if (unlikely(!pipe)) {
Jens Axboeb92ce552006-04-11 13:52:07 +02001021 pipe = alloc_pipe_info(NULL);
1022 if (!pipe)
1023 return -ENOMEM;
1024
1025 /*
1026 * We don't have an immediate reader, but we'll read the stuff
Jens Axboe00522fb2006-04-26 14:39:29 +02001027 * out of the pipe right after the splice_to_pipe(). So set
Jens Axboeb92ce552006-04-11 13:52:07 +02001028 * PIPE_READERS appropriately.
1029 */
1030 pipe->readers = 1;
1031
1032 current->splice_pipe = pipe;
1033 }
1034
1035 /*
Ingo Molnar73d62d82006-04-11 13:57:21 +02001036 * Do the splice.
Jens Axboeb92ce552006-04-11 13:52:07 +02001037 */
1038 ret = 0;
1039 bytes = 0;
Jens Axboecbb7e572006-04-11 14:57:50 +02001040 out_off = 0;
Jens Axboeb92ce552006-04-11 13:52:07 +02001041
1042 while (len) {
1043 size_t read_len, max_read_len;
1044
1045 /*
1046 * Do at most PIPE_BUFFERS pages worth of transfer:
1047 */
1048 max_read_len = min(len, (size_t)(PIPE_BUFFERS*PAGE_SIZE));
1049
Jens Axboecbb7e572006-04-11 14:57:50 +02001050 ret = do_splice_to(in, ppos, pipe, max_read_len, flags);
Jens Axboeb92ce552006-04-11 13:52:07 +02001051 if (unlikely(ret < 0))
1052 goto out_release;
1053
1054 read_len = ret;
1055
1056 /*
1057 * NOTE: nonblocking mode only applies to the input. We
1058 * must not do the output in nonblocking mode as then we
1059 * could get stuck data in the internal pipe:
1060 */
Jens Axboecbb7e572006-04-11 14:57:50 +02001061 ret = do_splice_from(pipe, out, &out_off, read_len,
Jens Axboeb92ce552006-04-11 13:52:07 +02001062 flags & ~SPLICE_F_NONBLOCK);
1063 if (unlikely(ret < 0))
1064 goto out_release;
1065
1066 bytes += ret;
1067 len -= ret;
1068
1069 /*
1070 * In nonblocking mode, if we got back a short read then
1071 * that was due to either an IO error or due to the
1072 * pagecache entry not being there. In the IO error case
1073 * the _next_ splice attempt will produce a clean IO error
1074 * return value (not a short read), so in both cases it's
1075 * correct to break out of the loop here:
1076 */
1077 if ((flags & SPLICE_F_NONBLOCK) && (read_len < max_read_len))
1078 break;
1079 }
1080
1081 pipe->nrbufs = pipe->curbuf = 0;
1082
1083 return bytes;
1084
1085out_release:
1086 /*
1087 * If we did an incomplete transfer we must release
1088 * the pipe buffers in question:
1089 */
1090 for (i = 0; i < PIPE_BUFFERS; i++) {
1091 struct pipe_buffer *buf = pipe->bufs + i;
1092
1093 if (buf->ops) {
1094 buf->ops->release(pipe, buf);
1095 buf->ops = NULL;
1096 }
1097 }
1098 pipe->nrbufs = pipe->curbuf = 0;
1099
1100 /*
1101 * If we transferred some data, return the number of bytes:
1102 */
1103 if (bytes > 0)
1104 return bytes;
1105
1106 return ret;
1107}
1108
1109EXPORT_SYMBOL(do_splice_direct);
1110
Jens Axboe83f91352006-04-02 23:05:09 +02001111/*
Jens Axboeddac0d32006-11-04 12:49:32 +01001112 * After the inode slimming patch, i_pipe/i_bdev/i_cdev share the same
1113 * location, so checking ->i_pipe is not enough to verify that this is a
1114 * pipe.
1115 */
1116static inline struct pipe_inode_info *pipe_info(struct inode *inode)
1117{
1118 if (S_ISFIFO(inode->i_mode))
1119 return inode->i_pipe;
1120
1121 return NULL;
1122}
1123
1124/*
Jens Axboe83f91352006-04-02 23:05:09 +02001125 * Determine where to splice to/from.
1126 */
Ingo Molnar529565d2006-04-10 15:18:58 +02001127static long do_splice(struct file *in, loff_t __user *off_in,
1128 struct file *out, loff_t __user *off_out,
1129 size_t len, unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +02001130{
Ingo Molnar3a326a22006-04-10 15:18:35 +02001131 struct pipe_inode_info *pipe;
Jens Axboecbb7e572006-04-11 14:57:50 +02001132 loff_t offset, *off;
Jens Axboea4514eb2006-04-19 15:57:05 +02001133 long ret;
Jens Axboe5274f052006-03-30 15:15:30 +02001134
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001135 pipe = pipe_info(in->f_path.dentry->d_inode);
Ingo Molnar529565d2006-04-10 15:18:58 +02001136 if (pipe) {
1137 if (off_in)
1138 return -ESPIPE;
Jens Axboeb92ce552006-04-11 13:52:07 +02001139 if (off_out) {
1140 if (out->f_op->llseek == no_llseek)
1141 return -EINVAL;
Jens Axboecbb7e572006-04-11 14:57:50 +02001142 if (copy_from_user(&offset, off_out, sizeof(loff_t)))
Jens Axboeb92ce552006-04-11 13:52:07 +02001143 return -EFAULT;
Jens Axboecbb7e572006-04-11 14:57:50 +02001144 off = &offset;
1145 } else
1146 off = &out->f_pos;
Ingo Molnar529565d2006-04-10 15:18:58 +02001147
Jens Axboea4514eb2006-04-19 15:57:05 +02001148 ret = do_splice_from(pipe, out, off, len, flags);
1149
1150 if (off_out && copy_to_user(off_out, off, sizeof(loff_t)))
1151 ret = -EFAULT;
1152
1153 return ret;
Ingo Molnar529565d2006-04-10 15:18:58 +02001154 }
Jens Axboe5274f052006-03-30 15:15:30 +02001155
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001156 pipe = pipe_info(out->f_path.dentry->d_inode);
Ingo Molnar529565d2006-04-10 15:18:58 +02001157 if (pipe) {
1158 if (off_out)
1159 return -ESPIPE;
Jens Axboeb92ce552006-04-11 13:52:07 +02001160 if (off_in) {
1161 if (in->f_op->llseek == no_llseek)
1162 return -EINVAL;
Jens Axboecbb7e572006-04-11 14:57:50 +02001163 if (copy_from_user(&offset, off_in, sizeof(loff_t)))
Jens Axboeb92ce552006-04-11 13:52:07 +02001164 return -EFAULT;
Jens Axboecbb7e572006-04-11 14:57:50 +02001165 off = &offset;
1166 } else
1167 off = &in->f_pos;
Ingo Molnar529565d2006-04-10 15:18:58 +02001168
Jens Axboea4514eb2006-04-19 15:57:05 +02001169 ret = do_splice_to(in, off, pipe, len, flags);
1170
1171 if (off_in && copy_to_user(off_in, off, sizeof(loff_t)))
1172 ret = -EFAULT;
1173
1174 return ret;
Ingo Molnar529565d2006-04-10 15:18:58 +02001175 }
Jens Axboe5274f052006-03-30 15:15:30 +02001176
1177 return -EINVAL;
1178}
1179
Jens Axboe912d35f2006-04-26 10:59:21 +02001180/*
1181 * Map an iov into an array of pages and offset/length tupples. With the
1182 * partial_page structure, we can map several non-contiguous ranges into
1183 * our ones pages[] map instead of splitting that operation into pieces.
1184 * Could easily be exported as a generic helper for other users, in which
1185 * case one would probably want to add a 'max_nr_pages' parameter as well.
1186 */
1187static int get_iovec_page_array(const struct iovec __user *iov,
1188 unsigned int nr_vecs, struct page **pages,
Jens Axboe7afa6fd2006-05-01 20:02:33 +02001189 struct partial_page *partial, int aligned)
Jens Axboe912d35f2006-04-26 10:59:21 +02001190{
1191 int buffers = 0, error = 0;
1192
1193 /*
1194 * It's ok to take the mmap_sem for reading, even
1195 * across a "get_user()".
1196 */
1197 down_read(&current->mm->mmap_sem);
1198
1199 while (nr_vecs) {
1200 unsigned long off, npages;
1201 void __user *base;
1202 size_t len;
1203 int i;
1204
1205 /*
1206 * Get user address base and length for this iovec.
1207 */
1208 error = get_user(base, &iov->iov_base);
1209 if (unlikely(error))
1210 break;
1211 error = get_user(len, &iov->iov_len);
1212 if (unlikely(error))
1213 break;
1214
1215 /*
1216 * Sanity check this iovec. 0 read succeeds.
1217 */
1218 if (unlikely(!len))
1219 break;
1220 error = -EFAULT;
1221 if (unlikely(!base))
1222 break;
1223
1224 /*
1225 * Get this base offset and number of pages, then map
1226 * in the user pages.
1227 */
1228 off = (unsigned long) base & ~PAGE_MASK;
Jens Axboe7afa6fd2006-05-01 20:02:33 +02001229
1230 /*
1231 * If asked for alignment, the offset must be zero and the
1232 * length a multiple of the PAGE_SIZE.
1233 */
1234 error = -EINVAL;
1235 if (aligned && (off || len & ~PAGE_MASK))
1236 break;
1237
Jens Axboe912d35f2006-04-26 10:59:21 +02001238 npages = (off + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1239 if (npages > PIPE_BUFFERS - buffers)
1240 npages = PIPE_BUFFERS - buffers;
1241
1242 error = get_user_pages(current, current->mm,
1243 (unsigned long) base, npages, 0, 0,
1244 &pages[buffers], NULL);
1245
1246 if (unlikely(error <= 0))
1247 break;
1248
1249 /*
1250 * Fill this contiguous range into the partial page map.
1251 */
1252 for (i = 0; i < error; i++) {
Jens Axboe75914892006-05-02 12:57:18 +02001253 const int plen = min_t(size_t, len, PAGE_SIZE - off);
Jens Axboe912d35f2006-04-26 10:59:21 +02001254
1255 partial[buffers].offset = off;
1256 partial[buffers].len = plen;
1257
1258 off = 0;
1259 len -= plen;
1260 buffers++;
1261 }
1262
1263 /*
1264 * We didn't complete this iov, stop here since it probably
1265 * means we have to move some of this into a pipe to
1266 * be able to continue.
1267 */
1268 if (len)
1269 break;
1270
1271 /*
1272 * Don't continue if we mapped fewer pages than we asked for,
1273 * or if we mapped the max number of pages that we have
1274 * room for.
1275 */
1276 if (error < npages || buffers == PIPE_BUFFERS)
1277 break;
1278
1279 nr_vecs--;
1280 iov++;
1281 }
1282
1283 up_read(&current->mm->mmap_sem);
1284
1285 if (buffers)
1286 return buffers;
1287
1288 return error;
1289}
1290
1291/*
1292 * vmsplice splices a user address range into a pipe. It can be thought of
1293 * as splice-from-memory, where the regular splice is splice-from-file (or
1294 * to file). In both cases the output is a pipe, naturally.
1295 *
1296 * Note that vmsplice only supports splicing _from_ user memory to a pipe,
1297 * not the other way around. Splicing from user memory is a simple operation
1298 * that can be supported without any funky alignment restrictions or nasty
1299 * vm tricks. We simply map in the user memory and fill them into a pipe.
1300 * The reverse isn't quite as easy, though. There are two possible solutions
1301 * for that:
1302 *
1303 * - memcpy() the data internally, at which point we might as well just
1304 * do a regular read() on the buffer anyway.
1305 * - Lots of nasty vm tricks, that are neither fast nor flexible (it
1306 * has restriction limitations on both ends of the pipe).
1307 *
1308 * Alas, it isn't here.
1309 *
1310 */
1311static long do_vmsplice(struct file *file, const struct iovec __user *iov,
1312 unsigned long nr_segs, unsigned int flags)
1313{
Jens Axboeddac0d32006-11-04 12:49:32 +01001314 struct pipe_inode_info *pipe;
Jens Axboe912d35f2006-04-26 10:59:21 +02001315 struct page *pages[PIPE_BUFFERS];
1316 struct partial_page partial[PIPE_BUFFERS];
1317 struct splice_pipe_desc spd = {
1318 .pages = pages,
1319 .partial = partial,
1320 .flags = flags,
1321 .ops = &user_page_pipe_buf_ops,
1322 };
1323
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001324 pipe = pipe_info(file->f_path.dentry->d_inode);
Jens Axboeddac0d32006-11-04 12:49:32 +01001325 if (!pipe)
Jens Axboe912d35f2006-04-26 10:59:21 +02001326 return -EBADF;
1327 if (unlikely(nr_segs > UIO_MAXIOV))
1328 return -EINVAL;
1329 else if (unlikely(!nr_segs))
1330 return 0;
1331
Jens Axboe7afa6fd2006-05-01 20:02:33 +02001332 spd.nr_pages = get_iovec_page_array(iov, nr_segs, pages, partial,
1333 flags & SPLICE_F_GIFT);
Jens Axboe912d35f2006-04-26 10:59:21 +02001334 if (spd.nr_pages <= 0)
1335 return spd.nr_pages;
1336
Jens Axboe00522fb2006-04-26 14:39:29 +02001337 return splice_to_pipe(pipe, &spd);
Jens Axboe912d35f2006-04-26 10:59:21 +02001338}
1339
1340asmlinkage long sys_vmsplice(int fd, const struct iovec __user *iov,
1341 unsigned long nr_segs, unsigned int flags)
1342{
1343 struct file *file;
1344 long error;
1345 int fput;
1346
1347 error = -EBADF;
1348 file = fget_light(fd, &fput);
1349 if (file) {
1350 if (file->f_mode & FMODE_WRITE)
1351 error = do_vmsplice(file, iov, nr_segs, flags);
1352
1353 fput_light(file, fput);
1354 }
1355
1356 return error;
1357}
1358
Ingo Molnar529565d2006-04-10 15:18:58 +02001359asmlinkage long sys_splice(int fd_in, loff_t __user *off_in,
1360 int fd_out, loff_t __user *off_out,
1361 size_t len, unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +02001362{
1363 long error;
1364 struct file *in, *out;
1365 int fput_in, fput_out;
1366
1367 if (unlikely(!len))
1368 return 0;
1369
1370 error = -EBADF;
Ingo Molnar529565d2006-04-10 15:18:58 +02001371 in = fget_light(fd_in, &fput_in);
Jens Axboe5274f052006-03-30 15:15:30 +02001372 if (in) {
1373 if (in->f_mode & FMODE_READ) {
Ingo Molnar529565d2006-04-10 15:18:58 +02001374 out = fget_light(fd_out, &fput_out);
Jens Axboe5274f052006-03-30 15:15:30 +02001375 if (out) {
1376 if (out->f_mode & FMODE_WRITE)
Ingo Molnar529565d2006-04-10 15:18:58 +02001377 error = do_splice(in, off_in,
1378 out, off_out,
1379 len, flags);
Jens Axboe5274f052006-03-30 15:15:30 +02001380 fput_light(out, fput_out);
1381 }
1382 }
1383
1384 fput_light(in, fput_in);
1385 }
1386
1387 return error;
1388}
Jens Axboe70524492006-04-11 15:51:17 +02001389
1390/*
Jens Axboeaadd06e2006-07-10 11:00:01 +02001391 * Make sure there's data to read. Wait for input if we can, otherwise
1392 * return an appropriate error.
1393 */
1394static int link_ipipe_prep(struct pipe_inode_info *pipe, unsigned int flags)
1395{
1396 int ret;
1397
1398 /*
1399 * Check ->nrbufs without the inode lock first. This function
1400 * is speculative anyways, so missing one is ok.
1401 */
1402 if (pipe->nrbufs)
1403 return 0;
1404
1405 ret = 0;
1406 mutex_lock(&pipe->inode->i_mutex);
1407
1408 while (!pipe->nrbufs) {
1409 if (signal_pending(current)) {
1410 ret = -ERESTARTSYS;
1411 break;
1412 }
1413 if (!pipe->writers)
1414 break;
1415 if (!pipe->waiting_writers) {
1416 if (flags & SPLICE_F_NONBLOCK) {
1417 ret = -EAGAIN;
1418 break;
1419 }
1420 }
1421 pipe_wait(pipe);
1422 }
1423
1424 mutex_unlock(&pipe->inode->i_mutex);
1425 return ret;
1426}
1427
1428/*
1429 * Make sure there's writeable room. Wait for room if we can, otherwise
1430 * return an appropriate error.
1431 */
1432static int link_opipe_prep(struct pipe_inode_info *pipe, unsigned int flags)
1433{
1434 int ret;
1435
1436 /*
1437 * Check ->nrbufs without the inode lock first. This function
1438 * is speculative anyways, so missing one is ok.
1439 */
1440 if (pipe->nrbufs < PIPE_BUFFERS)
1441 return 0;
1442
1443 ret = 0;
1444 mutex_lock(&pipe->inode->i_mutex);
1445
1446 while (pipe->nrbufs >= PIPE_BUFFERS) {
1447 if (!pipe->readers) {
1448 send_sig(SIGPIPE, current, 0);
1449 ret = -EPIPE;
1450 break;
1451 }
1452 if (flags & SPLICE_F_NONBLOCK) {
1453 ret = -EAGAIN;
1454 break;
1455 }
1456 if (signal_pending(current)) {
1457 ret = -ERESTARTSYS;
1458 break;
1459 }
1460 pipe->waiting_writers++;
1461 pipe_wait(pipe);
1462 pipe->waiting_writers--;
1463 }
1464
1465 mutex_unlock(&pipe->inode->i_mutex);
1466 return ret;
1467}
1468
1469/*
Jens Axboe70524492006-04-11 15:51:17 +02001470 * Link contents of ipipe to opipe.
1471 */
1472static int link_pipe(struct pipe_inode_info *ipipe,
1473 struct pipe_inode_info *opipe,
1474 size_t len, unsigned int flags)
1475{
1476 struct pipe_buffer *ibuf, *obuf;
Jens Axboeaadd06e2006-07-10 11:00:01 +02001477 int ret = 0, i = 0, nbuf;
Jens Axboe70524492006-04-11 15:51:17 +02001478
1479 /*
1480 * Potential ABBA deadlock, work around it by ordering lock
1481 * grabbing by inode address. Otherwise two different processes
1482 * could deadlock (one doing tee from A -> B, the other from B -> A).
1483 */
Mark Fasheh62752ee2006-10-17 10:31:38 +02001484 inode_double_lock(ipipe->inode, opipe->inode);
Jens Axboe70524492006-04-11 15:51:17 +02001485
Jens Axboeaadd06e2006-07-10 11:00:01 +02001486 do {
Jens Axboe70524492006-04-11 15:51:17 +02001487 if (!opipe->readers) {
1488 send_sig(SIGPIPE, current, 0);
1489 if (!ret)
1490 ret = -EPIPE;
1491 break;
1492 }
Jens Axboe70524492006-04-11 15:51:17 +02001493
1494 /*
Jens Axboeaadd06e2006-07-10 11:00:01 +02001495 * If we have iterated all input buffers or ran out of
1496 * output room, break.
Jens Axboe70524492006-04-11 15:51:17 +02001497 */
Jens Axboeaadd06e2006-07-10 11:00:01 +02001498 if (i >= ipipe->nrbufs || opipe->nrbufs >= PIPE_BUFFERS)
Jens Axboe70524492006-04-11 15:51:17 +02001499 break;
Jens Axboeaadd06e2006-07-10 11:00:01 +02001500
1501 ibuf = ipipe->bufs + ((ipipe->curbuf + i) & (PIPE_BUFFERS - 1));
1502 nbuf = (opipe->curbuf + opipe->nrbufs) & (PIPE_BUFFERS - 1);
1503
Jens Axboe2a27250e2006-04-19 15:56:40 +02001504 /*
Jens Axboeaadd06e2006-07-10 11:00:01 +02001505 * Get a reference to this pipe buffer,
1506 * so we can copy the contents over.
Jens Axboe2a27250e2006-04-19 15:56:40 +02001507 */
Jens Axboeaadd06e2006-07-10 11:00:01 +02001508 ibuf->ops->get(ipipe, ibuf);
Jens Axboe70524492006-04-11 15:51:17 +02001509
Jens Axboeaadd06e2006-07-10 11:00:01 +02001510 obuf = opipe->bufs + nbuf;
1511 *obuf = *ibuf;
Jens Axboe70524492006-04-11 15:51:17 +02001512
Jens Axboeaadd06e2006-07-10 11:00:01 +02001513 /*
1514 * Don't inherit the gift flag, we need to
1515 * prevent multiple steals of this page.
1516 */
1517 obuf->flags &= ~PIPE_BUF_FLAG_GIFT;
1518
1519 if (obuf->len > len)
1520 obuf->len = len;
1521
1522 opipe->nrbufs++;
1523 ret += obuf->len;
1524 len -= obuf->len;
1525 i++;
1526 } while (len);
Jens Axboe70524492006-04-11 15:51:17 +02001527
Mark Fasheh62752ee2006-10-17 10:31:38 +02001528 inode_double_unlock(ipipe->inode, opipe->inode);
Jens Axboe70524492006-04-11 15:51:17 +02001529
Jens Axboeaadd06e2006-07-10 11:00:01 +02001530 /*
1531 * If we put data in the output pipe, wakeup any potential readers.
1532 */
1533 if (ret > 0) {
Jens Axboe70524492006-04-11 15:51:17 +02001534 smp_mb();
1535 if (waitqueue_active(&opipe->wait))
1536 wake_up_interruptible(&opipe->wait);
1537 kill_fasync(&opipe->fasync_readers, SIGIO, POLL_IN);
1538 }
1539
1540 return ret;
1541}
1542
1543/*
1544 * This is a tee(1) implementation that works on pipes. It doesn't copy
1545 * any data, it simply references the 'in' pages on the 'out' pipe.
1546 * The 'flags' used are the SPLICE_F_* variants, currently the only
1547 * applicable one is SPLICE_F_NONBLOCK.
1548 */
1549static long do_tee(struct file *in, struct file *out, size_t len,
1550 unsigned int flags)
1551{
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001552 struct pipe_inode_info *ipipe = pipe_info(in->f_path.dentry->d_inode);
1553 struct pipe_inode_info *opipe = pipe_info(out->f_path.dentry->d_inode);
Jens Axboeaadd06e2006-07-10 11:00:01 +02001554 int ret = -EINVAL;
Jens Axboe70524492006-04-11 15:51:17 +02001555
1556 /*
Jens Axboeaadd06e2006-07-10 11:00:01 +02001557 * Duplicate the contents of ipipe to opipe without actually
1558 * copying the data.
Jens Axboe70524492006-04-11 15:51:17 +02001559 */
Jens Axboeaadd06e2006-07-10 11:00:01 +02001560 if (ipipe && opipe && ipipe != opipe) {
1561 /*
1562 * Keep going, unless we encounter an error. The ipipe/opipe
1563 * ordering doesn't really matter.
1564 */
1565 ret = link_ipipe_prep(ipipe, flags);
1566 if (!ret) {
1567 ret = link_opipe_prep(opipe, flags);
1568 if (!ret) {
1569 ret = link_pipe(ipipe, opipe, len, flags);
1570 if (!ret && (flags & SPLICE_F_NONBLOCK))
1571 ret = -EAGAIN;
1572 }
1573 }
1574 }
Jens Axboe70524492006-04-11 15:51:17 +02001575
Jens Axboeaadd06e2006-07-10 11:00:01 +02001576 return ret;
Jens Axboe70524492006-04-11 15:51:17 +02001577}
1578
1579asmlinkage long sys_tee(int fdin, int fdout, size_t len, unsigned int flags)
1580{
1581 struct file *in;
1582 int error, fput_in;
1583
1584 if (unlikely(!len))
1585 return 0;
1586
1587 error = -EBADF;
1588 in = fget_light(fdin, &fput_in);
1589 if (in) {
1590 if (in->f_mode & FMODE_READ) {
1591 int fput_out;
1592 struct file *out = fget_light(fdout, &fput_out);
1593
1594 if (out) {
1595 if (out->f_mode & FMODE_WRITE)
1596 error = do_tee(in, out, len, flags);
1597 fput_light(out, fput_out);
1598 }
1599 }
1600 fput_light(in, fput_in);
1601 }
1602
1603 return error;
1604}