15274f052SJens Axboe /* 25274f052SJens Axboe * "splice": joining two ropes together by interweaving their strands. 35274f052SJens Axboe * 45274f052SJens Axboe * This is the "extended pipe" functionality, where a pipe is used as 55274f052SJens Axboe * an arbitrary in-memory buffer. Think of a pipe as a small kernel 65274f052SJens Axboe * buffer that you can use to transfer data from one end to the other. 75274f052SJens Axboe * 85274f052SJens Axboe * The traditional unix read/write is extended with a "splice()" operation 95274f052SJens Axboe * that transfers data buffers to or from a pipe buffer. 105274f052SJens Axboe * 115274f052SJens Axboe * Named by Larry McVoy, original implementation from Linus, extended by 12c2058e06SJens Axboe * Jens to support splicing to files, network, direct splicing, etc and 13c2058e06SJens Axboe * fixing lots of bugs. 145274f052SJens Axboe * 15c2058e06SJens Axboe * Copyright (C) 2005-2006 Jens Axboe <axboe@suse.de> 16c2058e06SJens Axboe * Copyright (C) 2005-2006 Linus Torvalds <torvalds@osdl.org> 17c2058e06SJens Axboe * Copyright (C) 2006 Ingo Molnar <mingo@elte.hu> 185274f052SJens Axboe * 195274f052SJens Axboe */ 205274f052SJens Axboe #include <linux/fs.h> 215274f052SJens Axboe #include <linux/file.h> 225274f052SJens Axboe #include <linux/pagemap.h> 235274f052SJens Axboe #include <linux/pipe_fs_i.h> 245274f052SJens Axboe #include <linux/mm_inline.h> 255abc97aaSJens Axboe #include <linux/swap.h> 264f6f0bd2SJens Axboe #include <linux/writeback.h> 274f6f0bd2SJens Axboe #include <linux/buffer_head.h> 28a0f06780SJeff Garzik #include <linux/module.h> 294f6f0bd2SJens Axboe #include <linux/syscalls.h> 30912d35f8SJens Axboe #include <linux/uio.h> 315274f052SJens Axboe 32912d35f8SJens Axboe struct partial_page { 33912d35f8SJens Axboe unsigned int offset; 34912d35f8SJens Axboe unsigned int len; 35912d35f8SJens Axboe }; 36912d35f8SJens Axboe 37912d35f8SJens Axboe /* 3800522fb4SJens Axboe * Passed to splice_to_pipe 39912d35f8SJens Axboe */ 40912d35f8SJens Axboe struct splice_pipe_desc { 41912d35f8SJens Axboe struct page **pages; /* page map */ 42912d35f8SJens Axboe struct partial_page *partial; /* pages[] may not be contig */ 43912d35f8SJens Axboe int nr_pages; /* number of pages in map */ 44912d35f8SJens Axboe unsigned int flags; /* splice flags */ 45912d35f8SJens Axboe struct pipe_buf_operations *ops;/* ops associated with output pipe */ 46912d35f8SJens Axboe }; 47912d35f8SJens Axboe 4883f9135bSJens Axboe /* 4983f9135bSJens Axboe * Attempt to steal a page from a pipe buffer. This should perhaps go into 5083f9135bSJens Axboe * a vm helper function, it's already simplified quite a bit by the 5183f9135bSJens Axboe * addition of remove_mapping(). If success is returned, the caller may 5283f9135bSJens Axboe * attempt to reuse this page for another destination. 5383f9135bSJens Axboe */ 545abc97aaSJens Axboe static int page_cache_pipe_buf_steal(struct pipe_inode_info *info, 555abc97aaSJens Axboe struct pipe_buffer *buf) 565abc97aaSJens Axboe { 575abc97aaSJens Axboe struct page *page = buf->page; 584f6f0bd2SJens Axboe struct address_space *mapping = page_mapping(page); 595abc97aaSJens Axboe 609e0267c2SJens Axboe lock_page(page); 619e0267c2SJens Axboe 625abc97aaSJens Axboe WARN_ON(!PageUptodate(page)); 635abc97aaSJens Axboe 64ad8d6f0aSJens Axboe /* 65ad8d6f0aSJens Axboe * At least for ext2 with nobh option, we need to wait on writeback 66ad8d6f0aSJens Axboe * completing on this page, since we'll remove it from the pagecache. 67ad8d6f0aSJens Axboe * Otherwise truncate wont wait on the page, allowing the disk 68ad8d6f0aSJens Axboe * blocks to be reused by someone else before we actually wrote our 69ad8d6f0aSJens Axboe * data to them. fs corruption ensues. 70ad8d6f0aSJens Axboe */ 71ad8d6f0aSJens Axboe wait_on_page_writeback(page); 72ad8d6f0aSJens Axboe 734f6f0bd2SJens Axboe if (PagePrivate(page)) 744f6f0bd2SJens Axboe try_to_release_page(page, mapping_gfp_mask(mapping)); 754f6f0bd2SJens Axboe 769e0267c2SJens Axboe if (!remove_mapping(mapping, page)) { 779e0267c2SJens Axboe unlock_page(page); 785abc97aaSJens Axboe return 1; 799e0267c2SJens Axboe } 805abc97aaSJens Axboe 810568b409SJens Axboe buf->flags |= PIPE_BUF_FLAG_LRU; 825abc97aaSJens Axboe return 0; 835abc97aaSJens Axboe } 845abc97aaSJens Axboe 855274f052SJens Axboe static void page_cache_pipe_buf_release(struct pipe_inode_info *info, 865274f052SJens Axboe struct pipe_buffer *buf) 875274f052SJens Axboe { 885274f052SJens Axboe page_cache_release(buf->page); 895274f052SJens Axboe buf->page = NULL; 900568b409SJens Axboe buf->flags &= ~PIPE_BUF_FLAG_LRU; 915274f052SJens Axboe } 925274f052SJens Axboe 93f84d7519SJens Axboe static int page_cache_pipe_buf_pin(struct pipe_inode_info *info, 945274f052SJens Axboe struct pipe_buffer *buf) 955274f052SJens Axboe { 965274f052SJens Axboe struct page *page = buf->page; 9749d0b21bSJens Axboe int err; 985274f052SJens Axboe 995274f052SJens Axboe if (!PageUptodate(page)) { 10049d0b21bSJens Axboe lock_page(page); 1015274f052SJens Axboe 10249d0b21bSJens Axboe /* 10349d0b21bSJens Axboe * Page got truncated/unhashed. This will cause a 0-byte 10473d62d83SIngo Molnar * splice, if this is the first page. 10549d0b21bSJens Axboe */ 1065274f052SJens Axboe if (!page->mapping) { 10749d0b21bSJens Axboe err = -ENODATA; 10849d0b21bSJens Axboe goto error; 1095274f052SJens Axboe } 1105274f052SJens Axboe 11149d0b21bSJens Axboe /* 11273d62d83SIngo Molnar * Uh oh, read-error from disk. 11349d0b21bSJens Axboe */ 11449d0b21bSJens Axboe if (!PageUptodate(page)) { 11549d0b21bSJens Axboe err = -EIO; 11649d0b21bSJens Axboe goto error; 11749d0b21bSJens Axboe } 11849d0b21bSJens Axboe 11949d0b21bSJens Axboe /* 120f84d7519SJens Axboe * Page is ok afterall, we are done. 12149d0b21bSJens Axboe */ 12249d0b21bSJens Axboe unlock_page(page); 12349d0b21bSJens Axboe } 12449d0b21bSJens Axboe 125f84d7519SJens Axboe return 0; 12649d0b21bSJens Axboe error: 12749d0b21bSJens Axboe unlock_page(page); 128f84d7519SJens Axboe return err; 12970524490SJens Axboe } 13070524490SJens Axboe 1315274f052SJens Axboe static struct pipe_buf_operations page_cache_pipe_buf_ops = { 1325274f052SJens Axboe .can_merge = 0, 133f84d7519SJens Axboe .map = generic_pipe_buf_map, 134f84d7519SJens Axboe .unmap = generic_pipe_buf_unmap, 135f84d7519SJens Axboe .pin = page_cache_pipe_buf_pin, 1365274f052SJens Axboe .release = page_cache_pipe_buf_release, 1375abc97aaSJens Axboe .steal = page_cache_pipe_buf_steal, 138f84d7519SJens Axboe .get = generic_pipe_buf_get, 1395274f052SJens Axboe }; 1405274f052SJens Axboe 141912d35f8SJens Axboe static int user_page_pipe_buf_steal(struct pipe_inode_info *pipe, 142912d35f8SJens Axboe struct pipe_buffer *buf) 143912d35f8SJens Axboe { 1447afa6fd0SJens Axboe if (!(buf->flags & PIPE_BUF_FLAG_GIFT)) 145912d35f8SJens Axboe return 1; 1467afa6fd0SJens Axboe 1477afa6fd0SJens Axboe return 0; 148912d35f8SJens Axboe } 149912d35f8SJens Axboe 150912d35f8SJens Axboe static struct pipe_buf_operations user_page_pipe_buf_ops = { 151912d35f8SJens Axboe .can_merge = 0, 152f84d7519SJens Axboe .map = generic_pipe_buf_map, 153f84d7519SJens Axboe .unmap = generic_pipe_buf_unmap, 154f84d7519SJens Axboe .pin = generic_pipe_buf_pin, 155912d35f8SJens Axboe .release = page_cache_pipe_buf_release, 156912d35f8SJens Axboe .steal = user_page_pipe_buf_steal, 157f84d7519SJens Axboe .get = generic_pipe_buf_get, 158912d35f8SJens Axboe }; 159912d35f8SJens Axboe 16083f9135bSJens Axboe /* 16183f9135bSJens Axboe * Pipe output worker. This sets up our pipe format with the page cache 16283f9135bSJens Axboe * pipe buffer operations. Otherwise very similar to the regular pipe_writev(). 16383f9135bSJens Axboe */ 16400522fb4SJens Axboe static ssize_t splice_to_pipe(struct pipe_inode_info *pipe, 165912d35f8SJens Axboe struct splice_pipe_desc *spd) 1665274f052SJens Axboe { 167912d35f8SJens Axboe int ret, do_wakeup, page_nr; 1685274f052SJens Axboe 1695274f052SJens Axboe ret = 0; 1705274f052SJens Axboe do_wakeup = 0; 171912d35f8SJens Axboe page_nr = 0; 1725274f052SJens Axboe 1733a326a2cSIngo Molnar if (pipe->inode) 1743a326a2cSIngo Molnar mutex_lock(&pipe->inode->i_mutex); 1755274f052SJens Axboe 1765274f052SJens Axboe for (;;) { 1773a326a2cSIngo Molnar if (!pipe->readers) { 1785274f052SJens Axboe send_sig(SIGPIPE, current, 0); 1795274f052SJens Axboe if (!ret) 1805274f052SJens Axboe ret = -EPIPE; 1815274f052SJens Axboe break; 1825274f052SJens Axboe } 1835274f052SJens Axboe 1846f767b04SJens Axboe if (pipe->nrbufs < PIPE_BUFFERS) { 1856f767b04SJens Axboe int newbuf = (pipe->curbuf + pipe->nrbufs) & (PIPE_BUFFERS - 1); 1863a326a2cSIngo Molnar struct pipe_buffer *buf = pipe->bufs + newbuf; 1875274f052SJens Axboe 188912d35f8SJens Axboe buf->page = spd->pages[page_nr]; 189912d35f8SJens Axboe buf->offset = spd->partial[page_nr].offset; 190912d35f8SJens Axboe buf->len = spd->partial[page_nr].len; 191912d35f8SJens Axboe buf->ops = spd->ops; 1927afa6fd0SJens Axboe if (spd->flags & SPLICE_F_GIFT) 1937afa6fd0SJens Axboe buf->flags |= PIPE_BUF_FLAG_GIFT; 1947afa6fd0SJens Axboe 1956f767b04SJens Axboe pipe->nrbufs++; 196912d35f8SJens Axboe page_nr++; 197912d35f8SJens Axboe ret += buf->len; 198912d35f8SJens Axboe 1996f767b04SJens Axboe if (pipe->inode) 2005274f052SJens Axboe do_wakeup = 1; 2015274f052SJens Axboe 202912d35f8SJens Axboe if (!--spd->nr_pages) 2035274f052SJens Axboe break; 2046f767b04SJens Axboe if (pipe->nrbufs < PIPE_BUFFERS) 2055274f052SJens Axboe continue; 2065274f052SJens Axboe 2075274f052SJens Axboe break; 2085274f052SJens Axboe } 2095274f052SJens Axboe 210912d35f8SJens Axboe if (spd->flags & SPLICE_F_NONBLOCK) { 21129e35094SLinus Torvalds if (!ret) 21229e35094SLinus Torvalds ret = -EAGAIN; 21329e35094SLinus Torvalds break; 21429e35094SLinus Torvalds } 21529e35094SLinus Torvalds 2165274f052SJens Axboe if (signal_pending(current)) { 2175274f052SJens Axboe if (!ret) 2185274f052SJens Axboe ret = -ERESTARTSYS; 2195274f052SJens Axboe break; 2205274f052SJens Axboe } 2215274f052SJens Axboe 2225274f052SJens Axboe if (do_wakeup) { 223c0bd1f65SJens Axboe smp_mb(); 2243a326a2cSIngo Molnar if (waitqueue_active(&pipe->wait)) 2253a326a2cSIngo Molnar wake_up_interruptible_sync(&pipe->wait); 2263a326a2cSIngo Molnar kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN); 2275274f052SJens Axboe do_wakeup = 0; 2285274f052SJens Axboe } 2295274f052SJens Axboe 2303a326a2cSIngo Molnar pipe->waiting_writers++; 2313a326a2cSIngo Molnar pipe_wait(pipe); 2323a326a2cSIngo Molnar pipe->waiting_writers--; 2335274f052SJens Axboe } 2345274f052SJens Axboe 2353a326a2cSIngo Molnar if (pipe->inode) 2363a326a2cSIngo Molnar mutex_unlock(&pipe->inode->i_mutex); 2375274f052SJens Axboe 2385274f052SJens Axboe if (do_wakeup) { 239c0bd1f65SJens Axboe smp_mb(); 2403a326a2cSIngo Molnar if (waitqueue_active(&pipe->wait)) 2413a326a2cSIngo Molnar wake_up_interruptible(&pipe->wait); 2423a326a2cSIngo Molnar kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN); 2435274f052SJens Axboe } 2445274f052SJens Axboe 245912d35f8SJens Axboe while (page_nr < spd->nr_pages) 246912d35f8SJens Axboe page_cache_release(spd->pages[page_nr++]); 2475274f052SJens Axboe 2485274f052SJens Axboe return ret; 2495274f052SJens Axboe } 2505274f052SJens Axboe 2513a326a2cSIngo Molnar static int 252cbb7e577SJens Axboe __generic_file_splice_read(struct file *in, loff_t *ppos, 253cbb7e577SJens Axboe struct pipe_inode_info *pipe, size_t len, 254cbb7e577SJens Axboe unsigned int flags) 2555274f052SJens Axboe { 2565274f052SJens Axboe struct address_space *mapping = in->f_mapping; 257912d35f8SJens Axboe unsigned int loff, nr_pages; 25816c523ddSJens Axboe struct page *pages[PIPE_BUFFERS]; 259912d35f8SJens Axboe struct partial_page partial[PIPE_BUFFERS]; 2605274f052SJens Axboe struct page *page; 26191ad66efSJens Axboe pgoff_t index, end_index; 26291ad66efSJens Axboe loff_t isize; 263912d35f8SJens Axboe size_t total_len; 264eb20796bSJens Axboe int error, page_nr; 265912d35f8SJens Axboe struct splice_pipe_desc spd = { 266912d35f8SJens Axboe .pages = pages, 267912d35f8SJens Axboe .partial = partial, 268912d35f8SJens Axboe .flags = flags, 269912d35f8SJens Axboe .ops = &page_cache_pipe_buf_ops, 270912d35f8SJens Axboe }; 2715274f052SJens Axboe 272cbb7e577SJens Axboe index = *ppos >> PAGE_CACHE_SHIFT; 273912d35f8SJens Axboe loff = *ppos & ~PAGE_CACHE_MASK; 274912d35f8SJens Axboe nr_pages = (len + loff + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; 2755274f052SJens Axboe 2765274f052SJens Axboe if (nr_pages > PIPE_BUFFERS) 2775274f052SJens Axboe nr_pages = PIPE_BUFFERS; 2785274f052SJens Axboe 2795274f052SJens Axboe /* 28073d62d83SIngo Molnar * Initiate read-ahead on this page range. however, don't call into 2810b749ce3SJens Axboe * read-ahead if this is a non-zero offset (we are likely doing small 2820b749ce3SJens Axboe * chunk splice and the page is already there) for a single page. 2835274f052SJens Axboe */ 284eb645a24SJens Axboe if (!loff || nr_pages > 1) 285eb645a24SJens Axboe page_cache_readahead(mapping, &in->f_ra, in, index, nr_pages); 2865274f052SJens Axboe 2875274f052SJens Axboe /* 28873d62d83SIngo Molnar * Now fill in the holes: 2895274f052SJens Axboe */ 2907480a904SJens Axboe error = 0; 291912d35f8SJens Axboe total_len = 0; 29282aa5d61SJens Axboe 29382aa5d61SJens Axboe /* 294eb20796bSJens Axboe * Lookup the (hopefully) full range of pages we need. 29582aa5d61SJens Axboe */ 296eb20796bSJens Axboe spd.nr_pages = find_get_pages_contig(mapping, index, nr_pages, pages); 297eb20796bSJens Axboe 2985274f052SJens Axboe /* 299eb20796bSJens Axboe * If find_get_pages_contig() returned fewer pages than we needed, 300eb20796bSJens Axboe * allocate the rest. 301eb20796bSJens Axboe */ 302eb20796bSJens Axboe index += spd.nr_pages; 303eb20796bSJens Axboe while (spd.nr_pages < nr_pages) { 304eb20796bSJens Axboe /* 305eb20796bSJens Axboe * Page could be there, find_get_pages_contig() breaks on 306eb20796bSJens Axboe * the first hole. 3075274f052SJens Axboe */ 3087480a904SJens Axboe page = find_get_page(mapping, index); 3097480a904SJens Axboe if (!page) { 3107480a904SJens Axboe /* 311e27dedd8SJens Axboe * Make sure the read-ahead engine is notified 312e27dedd8SJens Axboe * about this failure. 313e27dedd8SJens Axboe */ 314e27dedd8SJens Axboe handle_ra_miss(mapping, &in->f_ra, index); 315e27dedd8SJens Axboe 316e27dedd8SJens Axboe /* 317eb20796bSJens Axboe * page didn't exist, allocate one. 3187480a904SJens Axboe */ 3197480a904SJens Axboe page = page_cache_alloc_cold(mapping); 3205274f052SJens Axboe if (!page) 3215274f052SJens Axboe break; 3225274f052SJens Axboe 3237480a904SJens Axboe error = add_to_page_cache_lru(page, mapping, index, 3247480a904SJens Axboe mapping_gfp_mask(mapping)); 3255274f052SJens Axboe if (unlikely(error)) { 3265274f052SJens Axboe page_cache_release(page); 3275274f052SJens Axboe break; 3285274f052SJens Axboe } 329eb20796bSJens Axboe /* 330eb20796bSJens Axboe * add_to_page_cache() locks the page, unlock it 331eb20796bSJens Axboe * to avoid convoluting the logic below even more. 332eb20796bSJens Axboe */ 333eb20796bSJens Axboe unlock_page(page); 3345274f052SJens Axboe } 3357480a904SJens Axboe 336eb20796bSJens Axboe pages[spd.nr_pages++] = page; 337eb20796bSJens Axboe index++; 338eb20796bSJens Axboe } 339eb20796bSJens Axboe 340eb20796bSJens Axboe /* 341eb20796bSJens Axboe * Now loop over the map and see if we need to start IO on any 342eb20796bSJens Axboe * pages, fill in the partial map, etc. 343eb20796bSJens Axboe */ 344eb20796bSJens Axboe index = *ppos >> PAGE_CACHE_SHIFT; 345eb20796bSJens Axboe nr_pages = spd.nr_pages; 346eb20796bSJens Axboe spd.nr_pages = 0; 347eb20796bSJens Axboe for (page_nr = 0; page_nr < nr_pages; page_nr++) { 348eb20796bSJens Axboe unsigned int this_len; 349eb20796bSJens Axboe 350eb20796bSJens Axboe if (!len) 351eb20796bSJens Axboe break; 352eb20796bSJens Axboe 353eb20796bSJens Axboe /* 354eb20796bSJens Axboe * this_len is the max we'll use from this page 355eb20796bSJens Axboe */ 356eb20796bSJens Axboe this_len = min_t(unsigned long, len, PAGE_CACHE_SIZE - loff); 357eb20796bSJens Axboe page = pages[page_nr]; 358eb20796bSJens Axboe 3597480a904SJens Axboe /* 3607480a904SJens Axboe * If the page isn't uptodate, we may need to start io on it 3617480a904SJens Axboe */ 3627480a904SJens Axboe if (!PageUptodate(page)) { 363c4f895cbSJens Axboe /* 364c4f895cbSJens Axboe * If in nonblock mode then dont block on waiting 365c4f895cbSJens Axboe * for an in-flight io page 366c4f895cbSJens Axboe */ 367c4f895cbSJens Axboe if (flags & SPLICE_F_NONBLOCK) 368c4f895cbSJens Axboe break; 369c4f895cbSJens Axboe 3707480a904SJens Axboe lock_page(page); 3717480a904SJens Axboe 3727480a904SJens Axboe /* 3737480a904SJens Axboe * page was truncated, stop here. if this isn't the 3747480a904SJens Axboe * first page, we'll just complete what we already 3757480a904SJens Axboe * added 3767480a904SJens Axboe */ 3777480a904SJens Axboe if (!page->mapping) { 3787480a904SJens Axboe unlock_page(page); 3797480a904SJens Axboe break; 3807480a904SJens Axboe } 3817480a904SJens Axboe /* 3827480a904SJens Axboe * page was already under io and is now done, great 3837480a904SJens Axboe */ 3847480a904SJens Axboe if (PageUptodate(page)) { 3857480a904SJens Axboe unlock_page(page); 3867480a904SJens Axboe goto fill_it; 3877480a904SJens Axboe } 3887480a904SJens Axboe 3897480a904SJens Axboe /* 3907480a904SJens Axboe * need to read in the page 3917480a904SJens Axboe */ 3927480a904SJens Axboe error = mapping->a_ops->readpage(in, page); 3937480a904SJens Axboe if (unlikely(error)) { 394eb20796bSJens Axboe /* 395eb20796bSJens Axboe * We really should re-lookup the page here, 396eb20796bSJens Axboe * but it complicates things a lot. Instead 397eb20796bSJens Axboe * lets just do what we already stored, and 398eb20796bSJens Axboe * we'll get it the next time we are called. 399eb20796bSJens Axboe */ 4007480a904SJens Axboe if (error == AOP_TRUNCATED_PAGE) 401eb20796bSJens Axboe error = 0; 402eb20796bSJens Axboe 4037480a904SJens Axboe break; 4047480a904SJens Axboe } 40591ad66efSJens Axboe 40691ad66efSJens Axboe /* 40791ad66efSJens Axboe * i_size must be checked after ->readpage(). 40891ad66efSJens Axboe */ 40991ad66efSJens Axboe isize = i_size_read(mapping->host); 41091ad66efSJens Axboe end_index = (isize - 1) >> PAGE_CACHE_SHIFT; 411eb20796bSJens Axboe if (unlikely(!isize || index > end_index)) 41291ad66efSJens Axboe break; 41391ad66efSJens Axboe 41491ad66efSJens Axboe /* 41591ad66efSJens Axboe * if this is the last page, see if we need to shrink 41691ad66efSJens Axboe * the length and stop 41791ad66efSJens Axboe */ 41891ad66efSJens Axboe if (end_index == index) { 41991ad66efSJens Axboe loff = PAGE_CACHE_SIZE - (isize & ~PAGE_CACHE_MASK); 420eb20796bSJens Axboe if (total_len + loff > isize) 42191ad66efSJens Axboe break; 42291ad66efSJens Axboe /* 42391ad66efSJens Axboe * force quit after adding this page 42491ad66efSJens Axboe */ 425eb20796bSJens Axboe len = this_len; 42682aa5d61SJens Axboe this_len = min(this_len, loff); 427912d35f8SJens Axboe loff = 0; 42891ad66efSJens Axboe } 4297480a904SJens Axboe } 4307480a904SJens Axboe fill_it: 431eb20796bSJens Axboe partial[page_nr].offset = loff; 432eb20796bSJens Axboe partial[page_nr].len = this_len; 43382aa5d61SJens Axboe len -= this_len; 434912d35f8SJens Axboe total_len += this_len; 43591ad66efSJens Axboe loff = 0; 436eb20796bSJens Axboe spd.nr_pages++; 437eb20796bSJens Axboe index++; 4385274f052SJens Axboe } 4395274f052SJens Axboe 440eb20796bSJens Axboe /* 441eb20796bSJens Axboe * Release any pages at the end, if we quit early. 'i' is how far 442eb20796bSJens Axboe * we got, 'nr_pages' is how many pages are in the map. 443eb20796bSJens Axboe */ 444eb20796bSJens Axboe while (page_nr < nr_pages) 445eb20796bSJens Axboe page_cache_release(pages[page_nr++]); 446eb20796bSJens Axboe 447912d35f8SJens Axboe if (spd.nr_pages) 44800522fb4SJens Axboe return splice_to_pipe(pipe, &spd); 44916c523ddSJens Axboe 4507480a904SJens Axboe return error; 4515274f052SJens Axboe } 4525274f052SJens Axboe 45383f9135bSJens Axboe /** 45483f9135bSJens Axboe * generic_file_splice_read - splice data from file to a pipe 45583f9135bSJens Axboe * @in: file to splice from 45683f9135bSJens Axboe * @pipe: pipe to splice to 45783f9135bSJens Axboe * @len: number of bytes to splice 45883f9135bSJens Axboe * @flags: splice modifier flags 45983f9135bSJens Axboe * 46083f9135bSJens Axboe * Will read pages from given file and fill them into a pipe. 46183f9135bSJens Axboe */ 462cbb7e577SJens Axboe ssize_t generic_file_splice_read(struct file *in, loff_t *ppos, 463cbb7e577SJens Axboe struct pipe_inode_info *pipe, size_t len, 464cbb7e577SJens Axboe unsigned int flags) 4655274f052SJens Axboe { 4665274f052SJens Axboe ssize_t spliced; 4675274f052SJens Axboe int ret; 4685274f052SJens Axboe 4695274f052SJens Axboe ret = 0; 4705274f052SJens Axboe spliced = 0; 4713a326a2cSIngo Molnar 4725274f052SJens Axboe while (len) { 473cbb7e577SJens Axboe ret = __generic_file_splice_read(in, ppos, pipe, len, flags); 4745274f052SJens Axboe 475c4f895cbSJens Axboe if (ret < 0) 4765274f052SJens Axboe break; 477c4f895cbSJens Axboe else if (!ret) { 478c4f895cbSJens Axboe if (spliced) 479c4f895cbSJens Axboe break; 480c4f895cbSJens Axboe if (flags & SPLICE_F_NONBLOCK) { 481c4f895cbSJens Axboe ret = -EAGAIN; 482c4f895cbSJens Axboe break; 483c4f895cbSJens Axboe } 484c4f895cbSJens Axboe } 4855274f052SJens Axboe 486cbb7e577SJens Axboe *ppos += ret; 4875274f052SJens Axboe len -= ret; 4885274f052SJens Axboe spliced += ret; 4895274f052SJens Axboe } 4905274f052SJens Axboe 4915274f052SJens Axboe if (spliced) 4925274f052SJens Axboe return spliced; 4935274f052SJens Axboe 4945274f052SJens Axboe return ret; 4955274f052SJens Axboe } 4965274f052SJens Axboe 497059a8f37SJens Axboe EXPORT_SYMBOL(generic_file_splice_read); 498059a8f37SJens Axboe 4995274f052SJens Axboe /* 5004f6f0bd2SJens Axboe * Send 'sd->len' bytes to socket from 'sd->file' at position 'sd->pos' 501016b661eSJens Axboe * using sendpage(). Return the number of bytes sent. 5025274f052SJens Axboe */ 5035274f052SJens Axboe static int pipe_to_sendpage(struct pipe_inode_info *info, 5045274f052SJens Axboe struct pipe_buffer *buf, struct splice_desc *sd) 5055274f052SJens Axboe { 5065274f052SJens Axboe struct file *file = sd->file; 5075274f052SJens Axboe loff_t pos = sd->pos; 508f84d7519SJens Axboe int ret, more; 5095274f052SJens Axboe 510f84d7519SJens Axboe ret = buf->ops->pin(info, buf); 511f84d7519SJens Axboe if (!ret) { 512b2b39fa4SJens Axboe more = (sd->flags & SPLICE_F_MORE) || sd->len < sd->total_len; 5135274f052SJens Axboe 514f84d7519SJens Axboe ret = file->f_op->sendpage(file, buf->page, buf->offset, 515f84d7519SJens Axboe sd->len, &pos, more); 516f84d7519SJens Axboe } 5175274f052SJens Axboe 518016b661eSJens Axboe return ret; 5195274f052SJens Axboe } 5205274f052SJens Axboe 5215274f052SJens Axboe /* 5225274f052SJens Axboe * This is a little more tricky than the file -> pipe splicing. There are 5235274f052SJens Axboe * basically three cases: 5245274f052SJens Axboe * 5255274f052SJens Axboe * - Destination page already exists in the address space and there 5265274f052SJens Axboe * are users of it. For that case we have no other option that 5275274f052SJens Axboe * copying the data. Tough luck. 5285274f052SJens Axboe * - Destination page already exists in the address space, but there 5295274f052SJens Axboe * are no users of it. Make sure it's uptodate, then drop it. Fall 5305274f052SJens Axboe * through to last case. 5315274f052SJens Axboe * - Destination page does not exist, we can add the pipe page to 5325274f052SJens Axboe * the page cache and avoid the copy. 5335274f052SJens Axboe * 53483f9135bSJens Axboe * If asked to move pages to the output file (SPLICE_F_MOVE is set in 53583f9135bSJens Axboe * sd->flags), we attempt to migrate pages from the pipe to the output 53683f9135bSJens Axboe * file address space page cache. This is possible if no one else has 53783f9135bSJens Axboe * the pipe page referenced outside of the pipe and page cache. If 53883f9135bSJens Axboe * SPLICE_F_MOVE isn't set, or we cannot move the page, we simply create 53983f9135bSJens Axboe * a new page in the output file page cache and fill/dirty that. 5405274f052SJens Axboe */ 5415274f052SJens Axboe static int pipe_to_file(struct pipe_inode_info *info, struct pipe_buffer *buf, 5425274f052SJens Axboe struct splice_desc *sd) 5435274f052SJens Axboe { 5445274f052SJens Axboe struct file *file = sd->file; 5455274f052SJens Axboe struct address_space *mapping = file->f_mapping; 5463e7ee3e7SJens Axboe gfp_t gfp_mask = mapping_gfp_mask(mapping); 547016b661eSJens Axboe unsigned int offset, this_len; 5485274f052SJens Axboe struct page *page; 5495274f052SJens Axboe pgoff_t index; 5503e7ee3e7SJens Axboe int ret; 5515274f052SJens Axboe 5525274f052SJens Axboe /* 55349d0b21bSJens Axboe * make sure the data in this buffer is uptodate 5545274f052SJens Axboe */ 555f84d7519SJens Axboe ret = buf->ops->pin(info, buf); 556f84d7519SJens Axboe if (unlikely(ret)) 557f84d7519SJens Axboe return ret; 5585274f052SJens Axboe 5595274f052SJens Axboe index = sd->pos >> PAGE_CACHE_SHIFT; 5605274f052SJens Axboe offset = sd->pos & ~PAGE_CACHE_MASK; 5615274f052SJens Axboe 562016b661eSJens Axboe this_len = sd->len; 563016b661eSJens Axboe if (this_len + offset > PAGE_CACHE_SIZE) 564016b661eSJens Axboe this_len = PAGE_CACHE_SIZE - offset; 565016b661eSJens Axboe 5665abc97aaSJens Axboe /* 5670568b409SJens Axboe * Reuse buf page, if SPLICE_F_MOVE is set and we are doing a full 5680568b409SJens Axboe * page. 5695abc97aaSJens Axboe */ 5700568b409SJens Axboe if ((sd->flags & SPLICE_F_MOVE) && this_len == PAGE_CACHE_SIZE) { 57183f9135bSJens Axboe /* 57283f9135bSJens Axboe * If steal succeeds, buf->page is now pruned from the vm 5739e0267c2SJens Axboe * side (LRU and page cache) and we can reuse it. The page 5749e0267c2SJens Axboe * will also be looked on successful return. 57583f9135bSJens Axboe */ 5765abc97aaSJens Axboe if (buf->ops->steal(info, buf)) 5775abc97aaSJens Axboe goto find_page; 5785abc97aaSJens Axboe 5795abc97aaSJens Axboe page = buf->page; 58046e678c9SJens Axboe if (add_to_page_cache(page, mapping, index, gfp_mask)) { 58146e678c9SJens Axboe unlock_page(page); 5825abc97aaSJens Axboe goto find_page; 58346e678c9SJens Axboe } 5843e7ee3e7SJens Axboe 5850568b409SJens Axboe page_cache_get(page); 5860568b409SJens Axboe 5873e7ee3e7SJens Axboe if (!(buf->flags & PIPE_BUF_FLAG_LRU)) 5883e7ee3e7SJens Axboe lru_cache_add(page); 5895abc97aaSJens Axboe } else { 5905274f052SJens Axboe find_page: 5919e0267c2SJens Axboe page = find_lock_page(mapping, index); 5929e0267c2SJens Axboe if (!page) { 5935274f052SJens Axboe ret = -ENOMEM; 5949e0267c2SJens Axboe page = page_cache_alloc_cold(mapping); 5959e0267c2SJens Axboe if (unlikely(!page)) 5969aefe431SDave Jones goto out_nomem; 5975274f052SJens Axboe 5985274f052SJens Axboe /* 5999e0267c2SJens Axboe * This will also lock the page 6009e0267c2SJens Axboe */ 6019e0267c2SJens Axboe ret = add_to_page_cache_lru(page, mapping, index, 6029e0267c2SJens Axboe gfp_mask); 6039e0267c2SJens Axboe if (unlikely(ret)) 6049e0267c2SJens Axboe goto out; 6059e0267c2SJens Axboe } 6069e0267c2SJens Axboe 6079e0267c2SJens Axboe /* 6089e0267c2SJens Axboe * We get here with the page locked. If the page is also 6099e0267c2SJens Axboe * uptodate, we don't need to do more. If it isn't, we 6109e0267c2SJens Axboe * may need to bring it in if we are not going to overwrite 6119e0267c2SJens Axboe * the full page. 6125274f052SJens Axboe */ 6135274f052SJens Axboe if (!PageUptodate(page)) { 614016b661eSJens Axboe if (this_len < PAGE_CACHE_SIZE) { 6155274f052SJens Axboe ret = mapping->a_ops->readpage(file, page); 6165274f052SJens Axboe if (unlikely(ret)) 6175274f052SJens Axboe goto out; 6185274f052SJens Axboe 6195274f052SJens Axboe lock_page(page); 6205274f052SJens Axboe 6215274f052SJens Axboe if (!PageUptodate(page)) { 6225274f052SJens Axboe /* 62373d62d83SIngo Molnar * Page got invalidated, repeat. 6245274f052SJens Axboe */ 6255274f052SJens Axboe if (!page->mapping) { 6265274f052SJens Axboe unlock_page(page); 6275274f052SJens Axboe page_cache_release(page); 6285274f052SJens Axboe goto find_page; 6295274f052SJens Axboe } 6305274f052SJens Axboe ret = -EIO; 6315274f052SJens Axboe goto out; 6325274f052SJens Axboe } 6339e0267c2SJens Axboe } else 6345274f052SJens Axboe SetPageUptodate(page); 6355274f052SJens Axboe } 6365274f052SJens Axboe } 6375274f052SJens Axboe 638016b661eSJens Axboe ret = mapping->a_ops->prepare_write(file, page, offset, offset+this_len); 6394f6f0bd2SJens Axboe if (ret == AOP_TRUNCATED_PAGE) { 6404f6f0bd2SJens Axboe page_cache_release(page); 6414f6f0bd2SJens Axboe goto find_page; 6424f6f0bd2SJens Axboe } else if (ret) 6435274f052SJens Axboe goto out; 6445274f052SJens Axboe 6450568b409SJens Axboe if (buf->page != page) { 646f84d7519SJens Axboe /* 647f84d7519SJens Axboe * Careful, ->map() uses KM_USER0! 648f84d7519SJens Axboe */ 649f6762b7aSJens Axboe char *src = buf->ops->map(info, buf, 1); 650f84d7519SJens Axboe char *dst = kmap_atomic(page, KM_USER1); 6515abc97aaSJens Axboe 652016b661eSJens Axboe memcpy(dst + offset, src + buf->offset, this_len); 6535274f052SJens Axboe flush_dcache_page(page); 654f84d7519SJens Axboe kunmap_atomic(dst, KM_USER1); 655f6762b7aSJens Axboe buf->ops->unmap(info, buf, src); 6565abc97aaSJens Axboe } 6575274f052SJens Axboe 658016b661eSJens Axboe ret = mapping->a_ops->commit_write(file, page, offset, offset+this_len); 6590568b409SJens Axboe if (!ret) { 660016b661eSJens Axboe /* 6610568b409SJens Axboe * Return the number of bytes written and mark page as 6620568b409SJens Axboe * accessed, we are now done! 663016b661eSJens Axboe */ 664016b661eSJens Axboe ret = this_len; 665c7f21e4fSJens Axboe mark_page_accessed(page); 6664f6f0bd2SJens Axboe balance_dirty_pages_ratelimited(mapping); 6670568b409SJens Axboe } else if (ret == AOP_TRUNCATED_PAGE) { 6685274f052SJens Axboe page_cache_release(page); 6690568b409SJens Axboe goto find_page; 6700568b409SJens Axboe } 6710568b409SJens Axboe out: 6720568b409SJens Axboe page_cache_release(page); 6734f6f0bd2SJens Axboe unlock_page(page); 6749aefe431SDave Jones out_nomem: 6755274f052SJens Axboe return ret; 6765274f052SJens Axboe } 6775274f052SJens Axboe 67883f9135bSJens Axboe /* 67983f9135bSJens Axboe * Pipe input worker. Most of this logic works like a regular pipe, the 68083f9135bSJens Axboe * key here is the 'actor' worker passed in that actually moves the data 68183f9135bSJens Axboe * to the wanted destination. See pipe_to_file/pipe_to_sendpage above. 68283f9135bSJens Axboe */ 68300522fb4SJens Axboe ssize_t splice_from_pipe(struct pipe_inode_info *pipe, struct file *out, 684cbb7e577SJens Axboe loff_t *ppos, size_t len, unsigned int flags, 6855274f052SJens Axboe splice_actor *actor) 6865274f052SJens Axboe { 6875274f052SJens Axboe int ret, do_wakeup, err; 6885274f052SJens Axboe struct splice_desc sd; 6895274f052SJens Axboe 6905274f052SJens Axboe ret = 0; 6915274f052SJens Axboe do_wakeup = 0; 6925274f052SJens Axboe 6935274f052SJens Axboe sd.total_len = len; 6945274f052SJens Axboe sd.flags = flags; 6955274f052SJens Axboe sd.file = out; 696cbb7e577SJens Axboe sd.pos = *ppos; 6975274f052SJens Axboe 6983a326a2cSIngo Molnar if (pipe->inode) 6993a326a2cSIngo Molnar mutex_lock(&pipe->inode->i_mutex); 7005274f052SJens Axboe 7015274f052SJens Axboe for (;;) { 7026f767b04SJens Axboe if (pipe->nrbufs) { 7036f767b04SJens Axboe struct pipe_buffer *buf = pipe->bufs + pipe->curbuf; 7045274f052SJens Axboe struct pipe_buf_operations *ops = buf->ops; 7055274f052SJens Axboe 7065274f052SJens Axboe sd.len = buf->len; 7075274f052SJens Axboe if (sd.len > sd.total_len) 7085274f052SJens Axboe sd.len = sd.total_len; 7095274f052SJens Axboe 7103a326a2cSIngo Molnar err = actor(pipe, buf, &sd); 711016b661eSJens Axboe if (err <= 0) { 7125274f052SJens Axboe if (!ret && err != -ENODATA) 7135274f052SJens Axboe ret = err; 7145274f052SJens Axboe 7155274f052SJens Axboe break; 7165274f052SJens Axboe } 7175274f052SJens Axboe 718016b661eSJens Axboe ret += err; 719016b661eSJens Axboe buf->offset += err; 720016b661eSJens Axboe buf->len -= err; 721016b661eSJens Axboe 722016b661eSJens Axboe sd.len -= err; 723016b661eSJens Axboe sd.pos += err; 724016b661eSJens Axboe sd.total_len -= err; 725016b661eSJens Axboe if (sd.len) 726016b661eSJens Axboe continue; 72773d62d83SIngo Molnar 7285274f052SJens Axboe if (!buf->len) { 7295274f052SJens Axboe buf->ops = NULL; 7303a326a2cSIngo Molnar ops->release(pipe, buf); 7316f767b04SJens Axboe pipe->curbuf = (pipe->curbuf + 1) & (PIPE_BUFFERS - 1); 7326f767b04SJens Axboe pipe->nrbufs--; 7336f767b04SJens Axboe if (pipe->inode) 7345274f052SJens Axboe do_wakeup = 1; 7355274f052SJens Axboe } 7365274f052SJens Axboe 7375274f052SJens Axboe if (!sd.total_len) 7385274f052SJens Axboe break; 7395274f052SJens Axboe } 7405274f052SJens Axboe 7416f767b04SJens Axboe if (pipe->nrbufs) 7425274f052SJens Axboe continue; 7433a326a2cSIngo Molnar if (!pipe->writers) 7445274f052SJens Axboe break; 7453a326a2cSIngo Molnar if (!pipe->waiting_writers) { 7465274f052SJens Axboe if (ret) 7475274f052SJens Axboe break; 7485274f052SJens Axboe } 7495274f052SJens Axboe 75029e35094SLinus Torvalds if (flags & SPLICE_F_NONBLOCK) { 75129e35094SLinus Torvalds if (!ret) 75229e35094SLinus Torvalds ret = -EAGAIN; 75329e35094SLinus Torvalds break; 75429e35094SLinus Torvalds } 75529e35094SLinus Torvalds 7565274f052SJens Axboe if (signal_pending(current)) { 7575274f052SJens Axboe if (!ret) 7585274f052SJens Axboe ret = -ERESTARTSYS; 7595274f052SJens Axboe break; 7605274f052SJens Axboe } 7615274f052SJens Axboe 7625274f052SJens Axboe if (do_wakeup) { 763c0bd1f65SJens Axboe smp_mb(); 7643a326a2cSIngo Molnar if (waitqueue_active(&pipe->wait)) 7653a326a2cSIngo Molnar wake_up_interruptible_sync(&pipe->wait); 7663a326a2cSIngo Molnar kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT); 7675274f052SJens Axboe do_wakeup = 0; 7685274f052SJens Axboe } 7695274f052SJens Axboe 7703a326a2cSIngo Molnar pipe_wait(pipe); 7715274f052SJens Axboe } 7725274f052SJens Axboe 7733a326a2cSIngo Molnar if (pipe->inode) 7743a326a2cSIngo Molnar mutex_unlock(&pipe->inode->i_mutex); 7755274f052SJens Axboe 7765274f052SJens Axboe if (do_wakeup) { 777c0bd1f65SJens Axboe smp_mb(); 7783a326a2cSIngo Molnar if (waitqueue_active(&pipe->wait)) 7793a326a2cSIngo Molnar wake_up_interruptible(&pipe->wait); 7803a326a2cSIngo Molnar kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT); 7815274f052SJens Axboe } 7825274f052SJens Axboe 7835274f052SJens Axboe return ret; 7845274f052SJens Axboe } 7855274f052SJens Axboe 78683f9135bSJens Axboe /** 78783f9135bSJens Axboe * generic_file_splice_write - splice data from a pipe to a file 7883a326a2cSIngo Molnar * @pipe: pipe info 78983f9135bSJens Axboe * @out: file to write to 79083f9135bSJens Axboe * @len: number of bytes to splice 79183f9135bSJens Axboe * @flags: splice modifier flags 79283f9135bSJens Axboe * 79383f9135bSJens Axboe * Will either move or copy pages (determined by @flags options) from 79483f9135bSJens Axboe * the given pipe inode to the given file. 79583f9135bSJens Axboe * 79683f9135bSJens Axboe */ 7973a326a2cSIngo Molnar ssize_t 7983a326a2cSIngo Molnar generic_file_splice_write(struct pipe_inode_info *pipe, struct file *out, 799cbb7e577SJens Axboe loff_t *ppos, size_t len, unsigned int flags) 8005274f052SJens Axboe { 8014f6f0bd2SJens Axboe struct address_space *mapping = out->f_mapping; 8023a326a2cSIngo Molnar ssize_t ret; 8033a326a2cSIngo Molnar 80400522fb4SJens Axboe ret = splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_file); 805a4514ebdSJens Axboe if (ret > 0) { 806a4514ebdSJens Axboe struct inode *inode = mapping->host; 807a4514ebdSJens Axboe 808a4514ebdSJens Axboe *ppos += ret; 8094f6f0bd2SJens Axboe 8104f6f0bd2SJens Axboe /* 811a4514ebdSJens Axboe * If file or inode is SYNC and we actually wrote some data, 812a4514ebdSJens Axboe * sync it. 8134f6f0bd2SJens Axboe */ 814a4514ebdSJens Axboe if (unlikely((out->f_flags & O_SYNC) || IS_SYNC(inode))) { 8154f6f0bd2SJens Axboe int err; 8164f6f0bd2SJens Axboe 8174f6f0bd2SJens Axboe mutex_lock(&inode->i_mutex); 818a4514ebdSJens Axboe err = generic_osync_inode(inode, mapping, 8194f6f0bd2SJens Axboe OSYNC_METADATA|OSYNC_DATA); 8204f6f0bd2SJens Axboe mutex_unlock(&inode->i_mutex); 8214f6f0bd2SJens Axboe 8224f6f0bd2SJens Axboe if (err) 8234f6f0bd2SJens Axboe ret = err; 8244f6f0bd2SJens Axboe } 825a4514ebdSJens Axboe } 8264f6f0bd2SJens Axboe 8274f6f0bd2SJens Axboe return ret; 8285274f052SJens Axboe } 8295274f052SJens Axboe 830059a8f37SJens Axboe EXPORT_SYMBOL(generic_file_splice_write); 831059a8f37SJens Axboe 83283f9135bSJens Axboe /** 83383f9135bSJens Axboe * generic_splice_sendpage - splice data from a pipe to a socket 83483f9135bSJens Axboe * @inode: pipe inode 83583f9135bSJens Axboe * @out: socket to write to 83683f9135bSJens Axboe * @len: number of bytes to splice 83783f9135bSJens Axboe * @flags: splice modifier flags 83883f9135bSJens Axboe * 83983f9135bSJens Axboe * Will send @len bytes from the pipe to a network socket. No data copying 84083f9135bSJens Axboe * is involved. 84183f9135bSJens Axboe * 84283f9135bSJens Axboe */ 8433a326a2cSIngo Molnar ssize_t generic_splice_sendpage(struct pipe_inode_info *pipe, struct file *out, 844cbb7e577SJens Axboe loff_t *ppos, size_t len, unsigned int flags) 8455274f052SJens Axboe { 84600522fb4SJens Axboe return splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_sendpage); 8475274f052SJens Axboe } 8485274f052SJens Axboe 849059a8f37SJens Axboe EXPORT_SYMBOL(generic_splice_sendpage); 850a0f06780SJeff Garzik 85183f9135bSJens Axboe /* 85283f9135bSJens Axboe * Attempt to initiate a splice from pipe to file. 85383f9135bSJens Axboe */ 8543a326a2cSIngo Molnar static long do_splice_from(struct pipe_inode_info *pipe, struct file *out, 855cbb7e577SJens Axboe loff_t *ppos, size_t len, unsigned int flags) 8565274f052SJens Axboe { 8575274f052SJens Axboe int ret; 8585274f052SJens Axboe 85949570e9bSJens Axboe if (unlikely(!out->f_op || !out->f_op->splice_write)) 8605274f052SJens Axboe return -EINVAL; 8615274f052SJens Axboe 86249570e9bSJens Axboe if (unlikely(!(out->f_mode & FMODE_WRITE))) 8635274f052SJens Axboe return -EBADF; 8645274f052SJens Axboe 865cbb7e577SJens Axboe ret = rw_verify_area(WRITE, out, ppos, len); 8665274f052SJens Axboe if (unlikely(ret < 0)) 8675274f052SJens Axboe return ret; 8685274f052SJens Axboe 869cbb7e577SJens Axboe return out->f_op->splice_write(pipe, out, ppos, len, flags); 8705274f052SJens Axboe } 8715274f052SJens Axboe 87283f9135bSJens Axboe /* 87383f9135bSJens Axboe * Attempt to initiate a splice from a file to a pipe. 87483f9135bSJens Axboe */ 875cbb7e577SJens Axboe static long do_splice_to(struct file *in, loff_t *ppos, 876cbb7e577SJens Axboe struct pipe_inode_info *pipe, size_t len, 877cbb7e577SJens Axboe unsigned int flags) 8785274f052SJens Axboe { 879cbb7e577SJens Axboe loff_t isize, left; 8805274f052SJens Axboe int ret; 8815274f052SJens Axboe 88249570e9bSJens Axboe if (unlikely(!in->f_op || !in->f_op->splice_read)) 8835274f052SJens Axboe return -EINVAL; 8845274f052SJens Axboe 88549570e9bSJens Axboe if (unlikely(!(in->f_mode & FMODE_READ))) 8865274f052SJens Axboe return -EBADF; 8875274f052SJens Axboe 888cbb7e577SJens Axboe ret = rw_verify_area(READ, in, ppos, len); 8895274f052SJens Axboe if (unlikely(ret < 0)) 8905274f052SJens Axboe return ret; 8915274f052SJens Axboe 8925274f052SJens Axboe isize = i_size_read(in->f_mapping->host); 893cbb7e577SJens Axboe if (unlikely(*ppos >= isize)) 8945274f052SJens Axboe return 0; 8955274f052SJens Axboe 896cbb7e577SJens Axboe left = isize - *ppos; 89749570e9bSJens Axboe if (unlikely(left < len)) 8985274f052SJens Axboe len = left; 8995274f052SJens Axboe 900cbb7e577SJens Axboe return in->f_op->splice_read(in, ppos, pipe, len, flags); 9015274f052SJens Axboe } 9025274f052SJens Axboe 903cbb7e577SJens Axboe long do_splice_direct(struct file *in, loff_t *ppos, struct file *out, 904cbb7e577SJens Axboe size_t len, unsigned int flags) 905b92ce558SJens Axboe { 906b92ce558SJens Axboe struct pipe_inode_info *pipe; 907b92ce558SJens Axboe long ret, bytes; 908cbb7e577SJens Axboe loff_t out_off; 909b92ce558SJens Axboe umode_t i_mode; 910b92ce558SJens Axboe int i; 911b92ce558SJens Axboe 912b92ce558SJens Axboe /* 913b92ce558SJens Axboe * We require the input being a regular file, as we don't want to 914b92ce558SJens Axboe * randomly drop data for eg socket -> socket splicing. Use the 915b92ce558SJens Axboe * piped splicing for that! 916b92ce558SJens Axboe */ 917b92ce558SJens Axboe i_mode = in->f_dentry->d_inode->i_mode; 918b92ce558SJens Axboe if (unlikely(!S_ISREG(i_mode) && !S_ISBLK(i_mode))) 919b92ce558SJens Axboe return -EINVAL; 920b92ce558SJens Axboe 921b92ce558SJens Axboe /* 922b92ce558SJens Axboe * neither in nor out is a pipe, setup an internal pipe attached to 923b92ce558SJens Axboe * 'out' and transfer the wanted data from 'in' to 'out' through that 924b92ce558SJens Axboe */ 925b92ce558SJens Axboe pipe = current->splice_pipe; 92649570e9bSJens Axboe if (unlikely(!pipe)) { 927b92ce558SJens Axboe pipe = alloc_pipe_info(NULL); 928b92ce558SJens Axboe if (!pipe) 929b92ce558SJens Axboe return -ENOMEM; 930b92ce558SJens Axboe 931b92ce558SJens Axboe /* 932b92ce558SJens Axboe * We don't have an immediate reader, but we'll read the stuff 93300522fb4SJens Axboe * out of the pipe right after the splice_to_pipe(). So set 934b92ce558SJens Axboe * PIPE_READERS appropriately. 935b92ce558SJens Axboe */ 936b92ce558SJens Axboe pipe->readers = 1; 937b92ce558SJens Axboe 938b92ce558SJens Axboe current->splice_pipe = pipe; 939b92ce558SJens Axboe } 940b92ce558SJens Axboe 941b92ce558SJens Axboe /* 94273d62d83SIngo Molnar * Do the splice. 943b92ce558SJens Axboe */ 944b92ce558SJens Axboe ret = 0; 945b92ce558SJens Axboe bytes = 0; 946cbb7e577SJens Axboe out_off = 0; 947b92ce558SJens Axboe 948b92ce558SJens Axboe while (len) { 949b92ce558SJens Axboe size_t read_len, max_read_len; 950b92ce558SJens Axboe 951b92ce558SJens Axboe /* 952b92ce558SJens Axboe * Do at most PIPE_BUFFERS pages worth of transfer: 953b92ce558SJens Axboe */ 954b92ce558SJens Axboe max_read_len = min(len, (size_t)(PIPE_BUFFERS*PAGE_SIZE)); 955b92ce558SJens Axboe 956cbb7e577SJens Axboe ret = do_splice_to(in, ppos, pipe, max_read_len, flags); 957b92ce558SJens Axboe if (unlikely(ret < 0)) 958b92ce558SJens Axboe goto out_release; 959b92ce558SJens Axboe 960b92ce558SJens Axboe read_len = ret; 961b92ce558SJens Axboe 962b92ce558SJens Axboe /* 963b92ce558SJens Axboe * NOTE: nonblocking mode only applies to the input. We 964b92ce558SJens Axboe * must not do the output in nonblocking mode as then we 965b92ce558SJens Axboe * could get stuck data in the internal pipe: 966b92ce558SJens Axboe */ 967cbb7e577SJens Axboe ret = do_splice_from(pipe, out, &out_off, read_len, 968b92ce558SJens Axboe flags & ~SPLICE_F_NONBLOCK); 969b92ce558SJens Axboe if (unlikely(ret < 0)) 970b92ce558SJens Axboe goto out_release; 971b92ce558SJens Axboe 972b92ce558SJens Axboe bytes += ret; 973b92ce558SJens Axboe len -= ret; 974b92ce558SJens Axboe 975b92ce558SJens Axboe /* 976b92ce558SJens Axboe * In nonblocking mode, if we got back a short read then 977b92ce558SJens Axboe * that was due to either an IO error or due to the 978b92ce558SJens Axboe * pagecache entry not being there. In the IO error case 979b92ce558SJens Axboe * the _next_ splice attempt will produce a clean IO error 980b92ce558SJens Axboe * return value (not a short read), so in both cases it's 981b92ce558SJens Axboe * correct to break out of the loop here: 982b92ce558SJens Axboe */ 983b92ce558SJens Axboe if ((flags & SPLICE_F_NONBLOCK) && (read_len < max_read_len)) 984b92ce558SJens Axboe break; 985b92ce558SJens Axboe } 986b92ce558SJens Axboe 987b92ce558SJens Axboe pipe->nrbufs = pipe->curbuf = 0; 988b92ce558SJens Axboe 989b92ce558SJens Axboe return bytes; 990b92ce558SJens Axboe 991b92ce558SJens Axboe out_release: 992b92ce558SJens Axboe /* 993b92ce558SJens Axboe * If we did an incomplete transfer we must release 994b92ce558SJens Axboe * the pipe buffers in question: 995b92ce558SJens Axboe */ 996b92ce558SJens Axboe for (i = 0; i < PIPE_BUFFERS; i++) { 997b92ce558SJens Axboe struct pipe_buffer *buf = pipe->bufs + i; 998b92ce558SJens Axboe 999b92ce558SJens Axboe if (buf->ops) { 1000b92ce558SJens Axboe buf->ops->release(pipe, buf); 1001b92ce558SJens Axboe buf->ops = NULL; 1002b92ce558SJens Axboe } 1003b92ce558SJens Axboe } 1004b92ce558SJens Axboe pipe->nrbufs = pipe->curbuf = 0; 1005b92ce558SJens Axboe 1006b92ce558SJens Axboe /* 1007b92ce558SJens Axboe * If we transferred some data, return the number of bytes: 1008b92ce558SJens Axboe */ 1009b92ce558SJens Axboe if (bytes > 0) 1010b92ce558SJens Axboe return bytes; 1011b92ce558SJens Axboe 1012b92ce558SJens Axboe return ret; 1013b92ce558SJens Axboe } 1014b92ce558SJens Axboe 1015b92ce558SJens Axboe EXPORT_SYMBOL(do_splice_direct); 1016b92ce558SJens Axboe 101783f9135bSJens Axboe /* 101883f9135bSJens Axboe * Determine where to splice to/from. 101983f9135bSJens Axboe */ 1020529565dcSIngo Molnar static long do_splice(struct file *in, loff_t __user *off_in, 1021529565dcSIngo Molnar struct file *out, loff_t __user *off_out, 1022529565dcSIngo Molnar size_t len, unsigned int flags) 10235274f052SJens Axboe { 10243a326a2cSIngo Molnar struct pipe_inode_info *pipe; 1025cbb7e577SJens Axboe loff_t offset, *off; 1026a4514ebdSJens Axboe long ret; 10275274f052SJens Axboe 10283a326a2cSIngo Molnar pipe = in->f_dentry->d_inode->i_pipe; 1029529565dcSIngo Molnar if (pipe) { 1030529565dcSIngo Molnar if (off_in) 1031529565dcSIngo Molnar return -ESPIPE; 1032b92ce558SJens Axboe if (off_out) { 1033b92ce558SJens Axboe if (out->f_op->llseek == no_llseek) 1034b92ce558SJens Axboe return -EINVAL; 1035cbb7e577SJens Axboe if (copy_from_user(&offset, off_out, sizeof(loff_t))) 1036b92ce558SJens Axboe return -EFAULT; 1037cbb7e577SJens Axboe off = &offset; 1038cbb7e577SJens Axboe } else 1039cbb7e577SJens Axboe off = &out->f_pos; 1040529565dcSIngo Molnar 1041a4514ebdSJens Axboe ret = do_splice_from(pipe, out, off, len, flags); 1042a4514ebdSJens Axboe 1043a4514ebdSJens Axboe if (off_out && copy_to_user(off_out, off, sizeof(loff_t))) 1044a4514ebdSJens Axboe ret = -EFAULT; 1045a4514ebdSJens Axboe 1046a4514ebdSJens Axboe return ret; 1047529565dcSIngo Molnar } 10485274f052SJens Axboe 10493a326a2cSIngo Molnar pipe = out->f_dentry->d_inode->i_pipe; 1050529565dcSIngo Molnar if (pipe) { 1051529565dcSIngo Molnar if (off_out) 1052529565dcSIngo Molnar return -ESPIPE; 1053b92ce558SJens Axboe if (off_in) { 1054b92ce558SJens Axboe if (in->f_op->llseek == no_llseek) 1055b92ce558SJens Axboe return -EINVAL; 1056cbb7e577SJens Axboe if (copy_from_user(&offset, off_in, sizeof(loff_t))) 1057b92ce558SJens Axboe return -EFAULT; 1058cbb7e577SJens Axboe off = &offset; 1059cbb7e577SJens Axboe } else 1060cbb7e577SJens Axboe off = &in->f_pos; 1061529565dcSIngo Molnar 1062a4514ebdSJens Axboe ret = do_splice_to(in, off, pipe, len, flags); 1063a4514ebdSJens Axboe 1064a4514ebdSJens Axboe if (off_in && copy_to_user(off_in, off, sizeof(loff_t))) 1065a4514ebdSJens Axboe ret = -EFAULT; 1066a4514ebdSJens Axboe 1067a4514ebdSJens Axboe return ret; 1068529565dcSIngo Molnar } 10695274f052SJens Axboe 10705274f052SJens Axboe return -EINVAL; 10715274f052SJens Axboe } 10725274f052SJens Axboe 1073912d35f8SJens Axboe /* 1074912d35f8SJens Axboe * Map an iov into an array of pages and offset/length tupples. With the 1075912d35f8SJens Axboe * partial_page structure, we can map several non-contiguous ranges into 1076912d35f8SJens Axboe * our ones pages[] map instead of splitting that operation into pieces. 1077912d35f8SJens Axboe * Could easily be exported as a generic helper for other users, in which 1078912d35f8SJens Axboe * case one would probably want to add a 'max_nr_pages' parameter as well. 1079912d35f8SJens Axboe */ 1080912d35f8SJens Axboe static int get_iovec_page_array(const struct iovec __user *iov, 1081912d35f8SJens Axboe unsigned int nr_vecs, struct page **pages, 10827afa6fd0SJens Axboe struct partial_page *partial, int aligned) 1083912d35f8SJens Axboe { 1084912d35f8SJens Axboe int buffers = 0, error = 0; 1085912d35f8SJens Axboe 1086912d35f8SJens Axboe /* 1087912d35f8SJens Axboe * It's ok to take the mmap_sem for reading, even 1088912d35f8SJens Axboe * across a "get_user()". 1089912d35f8SJens Axboe */ 1090912d35f8SJens Axboe down_read(¤t->mm->mmap_sem); 1091912d35f8SJens Axboe 1092912d35f8SJens Axboe while (nr_vecs) { 1093912d35f8SJens Axboe unsigned long off, npages; 1094912d35f8SJens Axboe void __user *base; 1095912d35f8SJens Axboe size_t len; 1096912d35f8SJens Axboe int i; 1097912d35f8SJens Axboe 1098912d35f8SJens Axboe /* 1099912d35f8SJens Axboe * Get user address base and length for this iovec. 1100912d35f8SJens Axboe */ 1101912d35f8SJens Axboe error = get_user(base, &iov->iov_base); 1102912d35f8SJens Axboe if (unlikely(error)) 1103912d35f8SJens Axboe break; 1104912d35f8SJens Axboe error = get_user(len, &iov->iov_len); 1105912d35f8SJens Axboe if (unlikely(error)) 1106912d35f8SJens Axboe break; 1107912d35f8SJens Axboe 1108912d35f8SJens Axboe /* 1109912d35f8SJens Axboe * Sanity check this iovec. 0 read succeeds. 1110912d35f8SJens Axboe */ 1111912d35f8SJens Axboe if (unlikely(!len)) 1112912d35f8SJens Axboe break; 1113912d35f8SJens Axboe error = -EFAULT; 1114912d35f8SJens Axboe if (unlikely(!base)) 1115912d35f8SJens Axboe break; 1116912d35f8SJens Axboe 1117912d35f8SJens Axboe /* 1118912d35f8SJens Axboe * Get this base offset and number of pages, then map 1119912d35f8SJens Axboe * in the user pages. 1120912d35f8SJens Axboe */ 1121912d35f8SJens Axboe off = (unsigned long) base & ~PAGE_MASK; 11227afa6fd0SJens Axboe 11237afa6fd0SJens Axboe /* 11247afa6fd0SJens Axboe * If asked for alignment, the offset must be zero and the 11257afa6fd0SJens Axboe * length a multiple of the PAGE_SIZE. 11267afa6fd0SJens Axboe */ 11277afa6fd0SJens Axboe error = -EINVAL; 11287afa6fd0SJens Axboe if (aligned && (off || len & ~PAGE_MASK)) 11297afa6fd0SJens Axboe break; 11307afa6fd0SJens Axboe 1131912d35f8SJens Axboe npages = (off + len + PAGE_SIZE - 1) >> PAGE_SHIFT; 1132912d35f8SJens Axboe if (npages > PIPE_BUFFERS - buffers) 1133912d35f8SJens Axboe npages = PIPE_BUFFERS - buffers; 1134912d35f8SJens Axboe 1135912d35f8SJens Axboe error = get_user_pages(current, current->mm, 1136912d35f8SJens Axboe (unsigned long) base, npages, 0, 0, 1137912d35f8SJens Axboe &pages[buffers], NULL); 1138912d35f8SJens Axboe 1139912d35f8SJens Axboe if (unlikely(error <= 0)) 1140912d35f8SJens Axboe break; 1141912d35f8SJens Axboe 1142912d35f8SJens Axboe /* 1143912d35f8SJens Axboe * Fill this contiguous range into the partial page map. 1144912d35f8SJens Axboe */ 1145912d35f8SJens Axboe for (i = 0; i < error; i++) { 11467591489aSJens Axboe const int plen = min_t(size_t, len, PAGE_SIZE - off); 1147912d35f8SJens Axboe 1148912d35f8SJens Axboe partial[buffers].offset = off; 1149912d35f8SJens Axboe partial[buffers].len = plen; 1150912d35f8SJens Axboe 1151912d35f8SJens Axboe off = 0; 1152912d35f8SJens Axboe len -= plen; 1153912d35f8SJens Axboe buffers++; 1154912d35f8SJens Axboe } 1155912d35f8SJens Axboe 1156912d35f8SJens Axboe /* 1157912d35f8SJens Axboe * We didn't complete this iov, stop here since it probably 1158912d35f8SJens Axboe * means we have to move some of this into a pipe to 1159912d35f8SJens Axboe * be able to continue. 1160912d35f8SJens Axboe */ 1161912d35f8SJens Axboe if (len) 1162912d35f8SJens Axboe break; 1163912d35f8SJens Axboe 1164912d35f8SJens Axboe /* 1165912d35f8SJens Axboe * Don't continue if we mapped fewer pages than we asked for, 1166912d35f8SJens Axboe * or if we mapped the max number of pages that we have 1167912d35f8SJens Axboe * room for. 1168912d35f8SJens Axboe */ 1169912d35f8SJens Axboe if (error < npages || buffers == PIPE_BUFFERS) 1170912d35f8SJens Axboe break; 1171912d35f8SJens Axboe 1172912d35f8SJens Axboe nr_vecs--; 1173912d35f8SJens Axboe iov++; 1174912d35f8SJens Axboe } 1175912d35f8SJens Axboe 1176912d35f8SJens Axboe up_read(¤t->mm->mmap_sem); 1177912d35f8SJens Axboe 1178912d35f8SJens Axboe if (buffers) 1179912d35f8SJens Axboe return buffers; 1180912d35f8SJens Axboe 1181912d35f8SJens Axboe return error; 1182912d35f8SJens Axboe } 1183912d35f8SJens Axboe 1184912d35f8SJens Axboe /* 1185912d35f8SJens Axboe * vmsplice splices a user address range into a pipe. It can be thought of 1186912d35f8SJens Axboe * as splice-from-memory, where the regular splice is splice-from-file (or 1187912d35f8SJens Axboe * to file). In both cases the output is a pipe, naturally. 1188912d35f8SJens Axboe * 1189912d35f8SJens Axboe * Note that vmsplice only supports splicing _from_ user memory to a pipe, 1190912d35f8SJens Axboe * not the other way around. Splicing from user memory is a simple operation 1191912d35f8SJens Axboe * that can be supported without any funky alignment restrictions or nasty 1192912d35f8SJens Axboe * vm tricks. We simply map in the user memory and fill them into a pipe. 1193912d35f8SJens Axboe * The reverse isn't quite as easy, though. There are two possible solutions 1194912d35f8SJens Axboe * for that: 1195912d35f8SJens Axboe * 1196912d35f8SJens Axboe * - memcpy() the data internally, at which point we might as well just 1197912d35f8SJens Axboe * do a regular read() on the buffer anyway. 1198912d35f8SJens Axboe * - Lots of nasty vm tricks, that are neither fast nor flexible (it 1199912d35f8SJens Axboe * has restriction limitations on both ends of the pipe). 1200912d35f8SJens Axboe * 1201912d35f8SJens Axboe * Alas, it isn't here. 1202912d35f8SJens Axboe * 1203912d35f8SJens Axboe */ 1204912d35f8SJens Axboe static long do_vmsplice(struct file *file, const struct iovec __user *iov, 1205912d35f8SJens Axboe unsigned long nr_segs, unsigned int flags) 1206912d35f8SJens Axboe { 1207912d35f8SJens Axboe struct pipe_inode_info *pipe = file->f_dentry->d_inode->i_pipe; 1208912d35f8SJens Axboe struct page *pages[PIPE_BUFFERS]; 1209912d35f8SJens Axboe struct partial_page partial[PIPE_BUFFERS]; 1210912d35f8SJens Axboe struct splice_pipe_desc spd = { 1211912d35f8SJens Axboe .pages = pages, 1212912d35f8SJens Axboe .partial = partial, 1213912d35f8SJens Axboe .flags = flags, 1214912d35f8SJens Axboe .ops = &user_page_pipe_buf_ops, 1215912d35f8SJens Axboe }; 1216912d35f8SJens Axboe 1217912d35f8SJens Axboe if (unlikely(!pipe)) 1218912d35f8SJens Axboe return -EBADF; 1219912d35f8SJens Axboe if (unlikely(nr_segs > UIO_MAXIOV)) 1220912d35f8SJens Axboe return -EINVAL; 1221912d35f8SJens Axboe else if (unlikely(!nr_segs)) 1222912d35f8SJens Axboe return 0; 1223912d35f8SJens Axboe 12247afa6fd0SJens Axboe spd.nr_pages = get_iovec_page_array(iov, nr_segs, pages, partial, 12257afa6fd0SJens Axboe flags & SPLICE_F_GIFT); 1226912d35f8SJens Axboe if (spd.nr_pages <= 0) 1227912d35f8SJens Axboe return spd.nr_pages; 1228912d35f8SJens Axboe 122900522fb4SJens Axboe return splice_to_pipe(pipe, &spd); 1230912d35f8SJens Axboe } 1231912d35f8SJens Axboe 1232912d35f8SJens Axboe asmlinkage long sys_vmsplice(int fd, const struct iovec __user *iov, 1233912d35f8SJens Axboe unsigned long nr_segs, unsigned int flags) 1234912d35f8SJens Axboe { 1235912d35f8SJens Axboe struct file *file; 1236912d35f8SJens Axboe long error; 1237912d35f8SJens Axboe int fput; 1238912d35f8SJens Axboe 1239912d35f8SJens Axboe error = -EBADF; 1240912d35f8SJens Axboe file = fget_light(fd, &fput); 1241912d35f8SJens Axboe if (file) { 1242912d35f8SJens Axboe if (file->f_mode & FMODE_WRITE) 1243912d35f8SJens Axboe error = do_vmsplice(file, iov, nr_segs, flags); 1244912d35f8SJens Axboe 1245912d35f8SJens Axboe fput_light(file, fput); 1246912d35f8SJens Axboe } 1247912d35f8SJens Axboe 1248912d35f8SJens Axboe return error; 1249912d35f8SJens Axboe } 1250912d35f8SJens Axboe 1251529565dcSIngo Molnar asmlinkage long sys_splice(int fd_in, loff_t __user *off_in, 1252529565dcSIngo Molnar int fd_out, loff_t __user *off_out, 1253529565dcSIngo Molnar size_t len, unsigned int flags) 12545274f052SJens Axboe { 12555274f052SJens Axboe long error; 12565274f052SJens Axboe struct file *in, *out; 12575274f052SJens Axboe int fput_in, fput_out; 12585274f052SJens Axboe 12595274f052SJens Axboe if (unlikely(!len)) 12605274f052SJens Axboe return 0; 12615274f052SJens Axboe 12625274f052SJens Axboe error = -EBADF; 1263529565dcSIngo Molnar in = fget_light(fd_in, &fput_in); 12645274f052SJens Axboe if (in) { 12655274f052SJens Axboe if (in->f_mode & FMODE_READ) { 1266529565dcSIngo Molnar out = fget_light(fd_out, &fput_out); 12675274f052SJens Axboe if (out) { 12685274f052SJens Axboe if (out->f_mode & FMODE_WRITE) 1269529565dcSIngo Molnar error = do_splice(in, off_in, 1270529565dcSIngo Molnar out, off_out, 1271529565dcSIngo Molnar len, flags); 12725274f052SJens Axboe fput_light(out, fput_out); 12735274f052SJens Axboe } 12745274f052SJens Axboe } 12755274f052SJens Axboe 12765274f052SJens Axboe fput_light(in, fput_in); 12775274f052SJens Axboe } 12785274f052SJens Axboe 12795274f052SJens Axboe return error; 12805274f052SJens Axboe } 128170524490SJens Axboe 128270524490SJens Axboe /* 128370524490SJens Axboe * Link contents of ipipe to opipe. 128470524490SJens Axboe */ 128570524490SJens Axboe static int link_pipe(struct pipe_inode_info *ipipe, 128670524490SJens Axboe struct pipe_inode_info *opipe, 128770524490SJens Axboe size_t len, unsigned int flags) 128870524490SJens Axboe { 128970524490SJens Axboe struct pipe_buffer *ibuf, *obuf; 12902a27250eSJens Axboe int ret, do_wakeup, i, ipipe_first; 12912a27250eSJens Axboe 12922a27250eSJens Axboe ret = do_wakeup = ipipe_first = 0; 129370524490SJens Axboe 129470524490SJens Axboe /* 129570524490SJens Axboe * Potential ABBA deadlock, work around it by ordering lock 129670524490SJens Axboe * grabbing by inode address. Otherwise two different processes 129770524490SJens Axboe * could deadlock (one doing tee from A -> B, the other from B -> A). 129870524490SJens Axboe */ 129970524490SJens Axboe if (ipipe->inode < opipe->inode) { 13002a27250eSJens Axboe ipipe_first = 1; 130170524490SJens Axboe mutex_lock(&ipipe->inode->i_mutex); 130270524490SJens Axboe mutex_lock(&opipe->inode->i_mutex); 130370524490SJens Axboe } else { 130470524490SJens Axboe mutex_lock(&opipe->inode->i_mutex); 130570524490SJens Axboe mutex_lock(&ipipe->inode->i_mutex); 130670524490SJens Axboe } 130770524490SJens Axboe 130870524490SJens Axboe for (i = 0;; i++) { 130970524490SJens Axboe if (!opipe->readers) { 131070524490SJens Axboe send_sig(SIGPIPE, current, 0); 131170524490SJens Axboe if (!ret) 131270524490SJens Axboe ret = -EPIPE; 131370524490SJens Axboe break; 131470524490SJens Axboe } 131570524490SJens Axboe if (ipipe->nrbufs - i) { 131670524490SJens Axboe ibuf = ipipe->bufs + ((ipipe->curbuf + i) & (PIPE_BUFFERS - 1)); 131770524490SJens Axboe 131870524490SJens Axboe /* 131970524490SJens Axboe * If we have room, fill this buffer 132070524490SJens Axboe */ 132170524490SJens Axboe if (opipe->nrbufs < PIPE_BUFFERS) { 132270524490SJens Axboe int nbuf = (opipe->curbuf + opipe->nrbufs) & (PIPE_BUFFERS - 1); 132370524490SJens Axboe 132470524490SJens Axboe /* 132570524490SJens Axboe * Get a reference to this pipe buffer, 132670524490SJens Axboe * so we can copy the contents over. 132770524490SJens Axboe */ 132870524490SJens Axboe ibuf->ops->get(ipipe, ibuf); 132970524490SJens Axboe 133070524490SJens Axboe obuf = opipe->bufs + nbuf; 133170524490SJens Axboe *obuf = *ibuf; 133270524490SJens Axboe 13337afa6fd0SJens Axboe /* 13347afa6fd0SJens Axboe * Don't inherit the gift flag, we need to 13357afa6fd0SJens Axboe * prevent multiple steals of this page. 13367afa6fd0SJens Axboe */ 13377afa6fd0SJens Axboe obuf->flags &= ~PIPE_BUF_FLAG_GIFT; 13387afa6fd0SJens Axboe 133970524490SJens Axboe if (obuf->len > len) 134070524490SJens Axboe obuf->len = len; 134170524490SJens Axboe 134270524490SJens Axboe opipe->nrbufs++; 134370524490SJens Axboe do_wakeup = 1; 134470524490SJens Axboe ret += obuf->len; 134570524490SJens Axboe len -= obuf->len; 134670524490SJens Axboe 134770524490SJens Axboe if (!len) 134870524490SJens Axboe break; 134970524490SJens Axboe if (opipe->nrbufs < PIPE_BUFFERS) 135070524490SJens Axboe continue; 135170524490SJens Axboe } 135270524490SJens Axboe 135370524490SJens Axboe /* 135470524490SJens Axboe * We have input available, but no output room. 13552a27250eSJens Axboe * If we already copied data, return that. If we 13562a27250eSJens Axboe * need to drop the opipe lock, it must be ordered 13572a27250eSJens Axboe * last to avoid deadlocks. 135870524490SJens Axboe */ 13592a27250eSJens Axboe if ((flags & SPLICE_F_NONBLOCK) || !ipipe_first) { 136070524490SJens Axboe if (!ret) 136170524490SJens Axboe ret = -EAGAIN; 136270524490SJens Axboe break; 136370524490SJens Axboe } 136470524490SJens Axboe if (signal_pending(current)) { 136570524490SJens Axboe if (!ret) 136670524490SJens Axboe ret = -ERESTARTSYS; 136770524490SJens Axboe break; 136870524490SJens Axboe } 136970524490SJens Axboe if (do_wakeup) { 137070524490SJens Axboe smp_mb(); 137170524490SJens Axboe if (waitqueue_active(&opipe->wait)) 137270524490SJens Axboe wake_up_interruptible(&opipe->wait); 137370524490SJens Axboe kill_fasync(&opipe->fasync_readers, SIGIO, POLL_IN); 137470524490SJens Axboe do_wakeup = 0; 137570524490SJens Axboe } 137670524490SJens Axboe 137770524490SJens Axboe opipe->waiting_writers++; 137870524490SJens Axboe pipe_wait(opipe); 137970524490SJens Axboe opipe->waiting_writers--; 138070524490SJens Axboe continue; 138170524490SJens Axboe } 138270524490SJens Axboe 138370524490SJens Axboe /* 138470524490SJens Axboe * No input buffers, do the usual checks for available 138570524490SJens Axboe * writers and blocking and wait if necessary 138670524490SJens Axboe */ 138770524490SJens Axboe if (!ipipe->writers) 138870524490SJens Axboe break; 138970524490SJens Axboe if (!ipipe->waiting_writers) { 139070524490SJens Axboe if (ret) 139170524490SJens Axboe break; 139270524490SJens Axboe } 13932a27250eSJens Axboe /* 13942a27250eSJens Axboe * pipe_wait() drops the ipipe mutex. To avoid deadlocks 13952a27250eSJens Axboe * with another process, we can only safely do that if 13962a27250eSJens Axboe * the ipipe lock is ordered last. 13972a27250eSJens Axboe */ 13982a27250eSJens Axboe if ((flags & SPLICE_F_NONBLOCK) || ipipe_first) { 139970524490SJens Axboe if (!ret) 140070524490SJens Axboe ret = -EAGAIN; 140170524490SJens Axboe break; 140270524490SJens Axboe } 140370524490SJens Axboe if (signal_pending(current)) { 140470524490SJens Axboe if (!ret) 140570524490SJens Axboe ret = -ERESTARTSYS; 140670524490SJens Axboe break; 140770524490SJens Axboe } 140870524490SJens Axboe 140970524490SJens Axboe if (waitqueue_active(&ipipe->wait)) 141070524490SJens Axboe wake_up_interruptible_sync(&ipipe->wait); 141170524490SJens Axboe kill_fasync(&ipipe->fasync_writers, SIGIO, POLL_OUT); 141270524490SJens Axboe 141370524490SJens Axboe pipe_wait(ipipe); 141470524490SJens Axboe } 141570524490SJens Axboe 141670524490SJens Axboe mutex_unlock(&ipipe->inode->i_mutex); 141770524490SJens Axboe mutex_unlock(&opipe->inode->i_mutex); 141870524490SJens Axboe 141970524490SJens Axboe if (do_wakeup) { 142070524490SJens Axboe smp_mb(); 142170524490SJens Axboe if (waitqueue_active(&opipe->wait)) 142270524490SJens Axboe wake_up_interruptible(&opipe->wait); 142370524490SJens Axboe kill_fasync(&opipe->fasync_readers, SIGIO, POLL_IN); 142470524490SJens Axboe } 142570524490SJens Axboe 142670524490SJens Axboe return ret; 142770524490SJens Axboe } 142870524490SJens Axboe 142970524490SJens Axboe /* 143070524490SJens Axboe * This is a tee(1) implementation that works on pipes. It doesn't copy 143170524490SJens Axboe * any data, it simply references the 'in' pages on the 'out' pipe. 143270524490SJens Axboe * The 'flags' used are the SPLICE_F_* variants, currently the only 143370524490SJens Axboe * applicable one is SPLICE_F_NONBLOCK. 143470524490SJens Axboe */ 143570524490SJens Axboe static long do_tee(struct file *in, struct file *out, size_t len, 143670524490SJens Axboe unsigned int flags) 143770524490SJens Axboe { 143870524490SJens Axboe struct pipe_inode_info *ipipe = in->f_dentry->d_inode->i_pipe; 143970524490SJens Axboe struct pipe_inode_info *opipe = out->f_dentry->d_inode->i_pipe; 144070524490SJens Axboe 144170524490SJens Axboe /* 144270524490SJens Axboe * Link ipipe to the two output pipes, consuming as we go along. 144370524490SJens Axboe */ 144470524490SJens Axboe if (ipipe && opipe) 144570524490SJens Axboe return link_pipe(ipipe, opipe, len, flags); 144670524490SJens Axboe 144770524490SJens Axboe return -EINVAL; 144870524490SJens Axboe } 144970524490SJens Axboe 145070524490SJens Axboe asmlinkage long sys_tee(int fdin, int fdout, size_t len, unsigned int flags) 145170524490SJens Axboe { 145270524490SJens Axboe struct file *in; 145370524490SJens Axboe int error, fput_in; 145470524490SJens Axboe 145570524490SJens Axboe if (unlikely(!len)) 145670524490SJens Axboe return 0; 145770524490SJens Axboe 145870524490SJens Axboe error = -EBADF; 145970524490SJens Axboe in = fget_light(fdin, &fput_in); 146070524490SJens Axboe if (in) { 146170524490SJens Axboe if (in->f_mode & FMODE_READ) { 146270524490SJens Axboe int fput_out; 146370524490SJens Axboe struct file *out = fget_light(fdout, &fput_out); 146470524490SJens Axboe 146570524490SJens Axboe if (out) { 146670524490SJens Axboe if (out->f_mode & FMODE_WRITE) 146770524490SJens Axboe error = do_tee(in, out, len, flags); 146870524490SJens Axboe fput_light(out, fput_out); 146970524490SJens Axboe } 147070524490SJens Axboe } 147170524490SJens Axboe fput_light(in, fput_in); 147270524490SJens Axboe } 147370524490SJens Axboe 147470524490SJens Axboe return error; 147570524490SJens Axboe } 1476