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 * 150fe23479SJens Axboe * Copyright (C) 2005-2006 Jens Axboe <axboe@kernel.dk> 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> 23d6b29d7cSJens Axboe #include <linux/splice.h> 2408e552c6SKAMEZAWA Hiroyuki #include <linux/memcontrol.h> 255274f052SJens Axboe #include <linux/mm_inline.h> 265abc97aaSJens Axboe #include <linux/swap.h> 274f6f0bd2SJens Axboe #include <linux/writeback.h> 28630d9c47SPaul Gortmaker #include <linux/export.h> 294f6f0bd2SJens Axboe #include <linux/syscalls.h> 30912d35f8SJens Axboe #include <linux/uio.h> 3129ce2058SJames Morris #include <linux/security.h> 325a0e3ad6STejun Heo #include <linux/gfp.h> 3335f9c09fSEric Dumazet #include <linux/socket.h> 3476b021d0SAl Viro #include <linux/compat.h> 3506ae43f3SAl Viro #include "internal.h" 365274f052SJens Axboe 3783f9135bSJens Axboe /* 3883f9135bSJens Axboe * Attempt to steal a page from a pipe buffer. This should perhaps go into 3983f9135bSJens Axboe * a vm helper function, it's already simplified quite a bit by the 4083f9135bSJens Axboe * addition of remove_mapping(). If success is returned, the caller may 4183f9135bSJens Axboe * attempt to reuse this page for another destination. 4283f9135bSJens Axboe */ 4376ad4d11SJens Axboe static int page_cache_pipe_buf_steal(struct pipe_inode_info *pipe, 445abc97aaSJens Axboe struct pipe_buffer *buf) 455abc97aaSJens Axboe { 465abc97aaSJens Axboe struct page *page = buf->page; 479e94cd4fSJens Axboe struct address_space *mapping; 485abc97aaSJens Axboe 499e0267c2SJens Axboe lock_page(page); 509e0267c2SJens Axboe 519e94cd4fSJens Axboe mapping = page_mapping(page); 529e94cd4fSJens Axboe if (mapping) { 535abc97aaSJens Axboe WARN_ON(!PageUptodate(page)); 545abc97aaSJens Axboe 55ad8d6f0aSJens Axboe /* 569e94cd4fSJens Axboe * At least for ext2 with nobh option, we need to wait on 579e94cd4fSJens Axboe * writeback completing on this page, since we'll remove it 589e94cd4fSJens Axboe * from the pagecache. Otherwise truncate wont wait on the 599e94cd4fSJens Axboe * page, allowing the disk blocks to be reused by someone else 609e94cd4fSJens Axboe * before we actually wrote our data to them. fs corruption 619e94cd4fSJens Axboe * ensues. 62ad8d6f0aSJens Axboe */ 63ad8d6f0aSJens Axboe wait_on_page_writeback(page); 64ad8d6f0aSJens Axboe 65266cf658SDavid Howells if (page_has_private(page) && 66266cf658SDavid Howells !try_to_release_page(page, GFP_KERNEL)) 67ca39d651SJens Axboe goto out_unlock; 684f6f0bd2SJens Axboe 699e94cd4fSJens Axboe /* 709e94cd4fSJens Axboe * If we succeeded in removing the mapping, set LRU flag 719e94cd4fSJens Axboe * and return good. 729e94cd4fSJens Axboe */ 739e94cd4fSJens Axboe if (remove_mapping(mapping, page)) { 741432873aSJens Axboe buf->flags |= PIPE_BUF_FLAG_LRU; 755abc97aaSJens Axboe return 0; 765abc97aaSJens Axboe } 779e94cd4fSJens Axboe } 789e94cd4fSJens Axboe 799e94cd4fSJens Axboe /* 809e94cd4fSJens Axboe * Raced with truncate or failed to remove page from current 819e94cd4fSJens Axboe * address space, unlock and return failure. 829e94cd4fSJens Axboe */ 83ca39d651SJens Axboe out_unlock: 849e94cd4fSJens Axboe unlock_page(page); 859e94cd4fSJens Axboe return 1; 869e94cd4fSJens Axboe } 875abc97aaSJens Axboe 8876ad4d11SJens Axboe static void page_cache_pipe_buf_release(struct pipe_inode_info *pipe, 895274f052SJens Axboe struct pipe_buffer *buf) 905274f052SJens Axboe { 915274f052SJens Axboe page_cache_release(buf->page); 921432873aSJens Axboe buf->flags &= ~PIPE_BUF_FLAG_LRU; 935274f052SJens Axboe } 945274f052SJens Axboe 950845718dSJens Axboe /* 960845718dSJens Axboe * Check whether the contents of buf is OK to access. Since the content 970845718dSJens Axboe * is a page cache page, IO may be in flight. 980845718dSJens Axboe */ 99cac36bb0SJens Axboe static int page_cache_pipe_buf_confirm(struct pipe_inode_info *pipe, 1005274f052SJens Axboe struct pipe_buffer *buf) 1015274f052SJens Axboe { 1025274f052SJens Axboe struct page *page = buf->page; 10349d0b21bSJens Axboe int err; 1045274f052SJens Axboe 1055274f052SJens Axboe if (!PageUptodate(page)) { 10649d0b21bSJens Axboe lock_page(page); 1075274f052SJens Axboe 10849d0b21bSJens Axboe /* 10949d0b21bSJens Axboe * Page got truncated/unhashed. This will cause a 0-byte 11073d62d83SIngo Molnar * splice, if this is the first page. 11149d0b21bSJens Axboe */ 1125274f052SJens Axboe if (!page->mapping) { 11349d0b21bSJens Axboe err = -ENODATA; 11449d0b21bSJens Axboe goto error; 1155274f052SJens Axboe } 1165274f052SJens Axboe 11749d0b21bSJens Axboe /* 11873d62d83SIngo Molnar * Uh oh, read-error from disk. 11949d0b21bSJens Axboe */ 12049d0b21bSJens Axboe if (!PageUptodate(page)) { 12149d0b21bSJens Axboe err = -EIO; 12249d0b21bSJens Axboe goto error; 12349d0b21bSJens Axboe } 12449d0b21bSJens Axboe 12549d0b21bSJens Axboe /* 126f84d7519SJens Axboe * Page is ok afterall, we are done. 12749d0b21bSJens Axboe */ 12849d0b21bSJens Axboe unlock_page(page); 12949d0b21bSJens Axboe } 13049d0b21bSJens Axboe 131f84d7519SJens Axboe return 0; 13249d0b21bSJens Axboe error: 13349d0b21bSJens Axboe unlock_page(page); 134f84d7519SJens Axboe return err; 13570524490SJens Axboe } 13670524490SJens Axboe 137708e3508SHugh Dickins const struct pipe_buf_operations page_cache_pipe_buf_ops = { 1385274f052SJens Axboe .can_merge = 0, 139f84d7519SJens Axboe .map = generic_pipe_buf_map, 140f84d7519SJens Axboe .unmap = generic_pipe_buf_unmap, 141cac36bb0SJens Axboe .confirm = page_cache_pipe_buf_confirm, 1425274f052SJens Axboe .release = page_cache_pipe_buf_release, 1435abc97aaSJens Axboe .steal = page_cache_pipe_buf_steal, 144f84d7519SJens Axboe .get = generic_pipe_buf_get, 1455274f052SJens Axboe }; 1465274f052SJens Axboe 147912d35f8SJens Axboe static int user_page_pipe_buf_steal(struct pipe_inode_info *pipe, 148912d35f8SJens Axboe struct pipe_buffer *buf) 149912d35f8SJens Axboe { 1507afa6fd0SJens Axboe if (!(buf->flags & PIPE_BUF_FLAG_GIFT)) 151912d35f8SJens Axboe return 1; 1527afa6fd0SJens Axboe 1531432873aSJens Axboe buf->flags |= PIPE_BUF_FLAG_LRU; 154330ab716SJens Axboe return generic_pipe_buf_steal(pipe, buf); 155912d35f8SJens Axboe } 156912d35f8SJens Axboe 157d4c3cca9SEric Dumazet static const struct pipe_buf_operations user_page_pipe_buf_ops = { 158912d35f8SJens Axboe .can_merge = 0, 159f84d7519SJens Axboe .map = generic_pipe_buf_map, 160f84d7519SJens Axboe .unmap = generic_pipe_buf_unmap, 161cac36bb0SJens Axboe .confirm = generic_pipe_buf_confirm, 162912d35f8SJens Axboe .release = page_cache_pipe_buf_release, 163912d35f8SJens Axboe .steal = user_page_pipe_buf_steal, 164f84d7519SJens Axboe .get = generic_pipe_buf_get, 165912d35f8SJens Axboe }; 166912d35f8SJens Axboe 167825cdcb1SNamhyung Kim static void wakeup_pipe_readers(struct pipe_inode_info *pipe) 168825cdcb1SNamhyung Kim { 169825cdcb1SNamhyung Kim smp_mb(); 170825cdcb1SNamhyung Kim if (waitqueue_active(&pipe->wait)) 171825cdcb1SNamhyung Kim wake_up_interruptible(&pipe->wait); 172825cdcb1SNamhyung Kim kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN); 173825cdcb1SNamhyung Kim } 174825cdcb1SNamhyung Kim 175932cc6d4SJens Axboe /** 176932cc6d4SJens Axboe * splice_to_pipe - fill passed data into a pipe 177932cc6d4SJens Axboe * @pipe: pipe to fill 178932cc6d4SJens Axboe * @spd: data to fill 179932cc6d4SJens Axboe * 180932cc6d4SJens Axboe * Description: 18179685b8dSRandy Dunlap * @spd contains a map of pages and len/offset tuples, along with 182932cc6d4SJens Axboe * the struct pipe_buf_operations associated with these pages. This 183932cc6d4SJens Axboe * function will link that data to the pipe. 184932cc6d4SJens Axboe * 18583f9135bSJens Axboe */ 186d6b29d7cSJens Axboe ssize_t splice_to_pipe(struct pipe_inode_info *pipe, 187912d35f8SJens Axboe struct splice_pipe_desc *spd) 1885274f052SJens Axboe { 18900de00bdSJens Axboe unsigned int spd_pages = spd->nr_pages; 190912d35f8SJens Axboe int ret, do_wakeup, page_nr; 1915274f052SJens Axboe 1925274f052SJens Axboe ret = 0; 1935274f052SJens Axboe do_wakeup = 0; 194912d35f8SJens Axboe page_nr = 0; 1955274f052SJens Axboe 19661e0d47cSMiklos Szeredi pipe_lock(pipe); 1975274f052SJens Axboe 1985274f052SJens Axboe for (;;) { 1993a326a2cSIngo Molnar if (!pipe->readers) { 2005274f052SJens Axboe send_sig(SIGPIPE, current, 0); 2015274f052SJens Axboe if (!ret) 2025274f052SJens Axboe ret = -EPIPE; 2035274f052SJens Axboe break; 2045274f052SJens Axboe } 2055274f052SJens Axboe 20635f3d14dSJens Axboe if (pipe->nrbufs < pipe->buffers) { 20735f3d14dSJens Axboe int newbuf = (pipe->curbuf + pipe->nrbufs) & (pipe->buffers - 1); 2083a326a2cSIngo Molnar struct pipe_buffer *buf = pipe->bufs + newbuf; 2095274f052SJens Axboe 210912d35f8SJens Axboe buf->page = spd->pages[page_nr]; 211912d35f8SJens Axboe buf->offset = spd->partial[page_nr].offset; 212912d35f8SJens Axboe buf->len = spd->partial[page_nr].len; 213497f9625SJens Axboe buf->private = spd->partial[page_nr].private; 214912d35f8SJens Axboe buf->ops = spd->ops; 2157afa6fd0SJens Axboe if (spd->flags & SPLICE_F_GIFT) 2167afa6fd0SJens Axboe buf->flags |= PIPE_BUF_FLAG_GIFT; 2177afa6fd0SJens Axboe 2186f767b04SJens Axboe pipe->nrbufs++; 219912d35f8SJens Axboe page_nr++; 220912d35f8SJens Axboe ret += buf->len; 221912d35f8SJens Axboe 2226447a3cfSAl Viro if (pipe->files) 2235274f052SJens Axboe do_wakeup = 1; 2245274f052SJens Axboe 225912d35f8SJens Axboe if (!--spd->nr_pages) 2265274f052SJens Axboe break; 22735f3d14dSJens Axboe if (pipe->nrbufs < pipe->buffers) 2285274f052SJens Axboe continue; 2295274f052SJens Axboe 2305274f052SJens Axboe break; 2315274f052SJens Axboe } 2325274f052SJens Axboe 233912d35f8SJens Axboe if (spd->flags & SPLICE_F_NONBLOCK) { 23429e35094SLinus Torvalds if (!ret) 23529e35094SLinus Torvalds ret = -EAGAIN; 23629e35094SLinus Torvalds break; 23729e35094SLinus Torvalds } 23829e35094SLinus Torvalds 2395274f052SJens Axboe if (signal_pending(current)) { 2405274f052SJens Axboe if (!ret) 2415274f052SJens Axboe ret = -ERESTARTSYS; 2425274f052SJens Axboe break; 2435274f052SJens Axboe } 2445274f052SJens Axboe 2455274f052SJens Axboe if (do_wakeup) { 246c0bd1f65SJens Axboe smp_mb(); 2473a326a2cSIngo Molnar if (waitqueue_active(&pipe->wait)) 2483a326a2cSIngo Molnar wake_up_interruptible_sync(&pipe->wait); 2493a326a2cSIngo Molnar kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN); 2505274f052SJens Axboe do_wakeup = 0; 2515274f052SJens Axboe } 2525274f052SJens Axboe 2533a326a2cSIngo Molnar pipe->waiting_writers++; 2543a326a2cSIngo Molnar pipe_wait(pipe); 2553a326a2cSIngo Molnar pipe->waiting_writers--; 2565274f052SJens Axboe } 2575274f052SJens Axboe 25861e0d47cSMiklos Szeredi pipe_unlock(pipe); 2595274f052SJens Axboe 260825cdcb1SNamhyung Kim if (do_wakeup) 261825cdcb1SNamhyung Kim wakeup_pipe_readers(pipe); 2625274f052SJens Axboe 26300de00bdSJens Axboe while (page_nr < spd_pages) 264bbdfc2f7SJens Axboe spd->spd_release(spd, page_nr++); 2655274f052SJens Axboe 2665274f052SJens Axboe return ret; 2675274f052SJens Axboe } 2685274f052SJens Axboe 269708e3508SHugh Dickins void spd_release_page(struct splice_pipe_desc *spd, unsigned int i) 270bbdfc2f7SJens Axboe { 271bbdfc2f7SJens Axboe page_cache_release(spd->pages[i]); 272bbdfc2f7SJens Axboe } 273bbdfc2f7SJens Axboe 27435f3d14dSJens Axboe /* 27535f3d14dSJens Axboe * Check if we need to grow the arrays holding pages and partial page 27635f3d14dSJens Axboe * descriptions. 27735f3d14dSJens Axboe */ 278047fe360SEric Dumazet int splice_grow_spd(const struct pipe_inode_info *pipe, struct splice_pipe_desc *spd) 27935f3d14dSJens Axboe { 280047fe360SEric Dumazet unsigned int buffers = ACCESS_ONCE(pipe->buffers); 281047fe360SEric Dumazet 282047fe360SEric Dumazet spd->nr_pages_max = buffers; 283047fe360SEric Dumazet if (buffers <= PIPE_DEF_BUFFERS) 28435f3d14dSJens Axboe return 0; 28535f3d14dSJens Axboe 286047fe360SEric Dumazet spd->pages = kmalloc(buffers * sizeof(struct page *), GFP_KERNEL); 287047fe360SEric Dumazet spd->partial = kmalloc(buffers * sizeof(struct partial_page), GFP_KERNEL); 28835f3d14dSJens Axboe 28935f3d14dSJens Axboe if (spd->pages && spd->partial) 29035f3d14dSJens Axboe return 0; 29135f3d14dSJens Axboe 29235f3d14dSJens Axboe kfree(spd->pages); 29335f3d14dSJens Axboe kfree(spd->partial); 29435f3d14dSJens Axboe return -ENOMEM; 29535f3d14dSJens Axboe } 29635f3d14dSJens Axboe 297047fe360SEric Dumazet void splice_shrink_spd(struct splice_pipe_desc *spd) 29835f3d14dSJens Axboe { 299047fe360SEric Dumazet if (spd->nr_pages_max <= PIPE_DEF_BUFFERS) 30035f3d14dSJens Axboe return; 30135f3d14dSJens Axboe 30235f3d14dSJens Axboe kfree(spd->pages); 30335f3d14dSJens Axboe kfree(spd->partial); 30435f3d14dSJens Axboe } 30535f3d14dSJens Axboe 3063a326a2cSIngo Molnar static int 307cbb7e577SJens Axboe __generic_file_splice_read(struct file *in, loff_t *ppos, 308cbb7e577SJens Axboe struct pipe_inode_info *pipe, size_t len, 309cbb7e577SJens Axboe unsigned int flags) 3105274f052SJens Axboe { 3115274f052SJens Axboe struct address_space *mapping = in->f_mapping; 312d8983910SFengguang Wu unsigned int loff, nr_pages, req_pages; 31335f3d14dSJens Axboe struct page *pages[PIPE_DEF_BUFFERS]; 31435f3d14dSJens Axboe struct partial_page partial[PIPE_DEF_BUFFERS]; 3155274f052SJens Axboe struct page *page; 31691ad66efSJens Axboe pgoff_t index, end_index; 31791ad66efSJens Axboe loff_t isize; 318eb20796bSJens Axboe int error, page_nr; 319912d35f8SJens Axboe struct splice_pipe_desc spd = { 320912d35f8SJens Axboe .pages = pages, 321912d35f8SJens Axboe .partial = partial, 322047fe360SEric Dumazet .nr_pages_max = PIPE_DEF_BUFFERS, 323912d35f8SJens Axboe .flags = flags, 324912d35f8SJens Axboe .ops = &page_cache_pipe_buf_ops, 325bbdfc2f7SJens Axboe .spd_release = spd_release_page, 326912d35f8SJens Axboe }; 3275274f052SJens Axboe 32835f3d14dSJens Axboe if (splice_grow_spd(pipe, &spd)) 32935f3d14dSJens Axboe return -ENOMEM; 33035f3d14dSJens Axboe 331cbb7e577SJens Axboe index = *ppos >> PAGE_CACHE_SHIFT; 332912d35f8SJens Axboe loff = *ppos & ~PAGE_CACHE_MASK; 333d8983910SFengguang Wu req_pages = (len + loff + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; 334047fe360SEric Dumazet nr_pages = min(req_pages, spd.nr_pages_max); 3355274f052SJens Axboe 3365274f052SJens Axboe /* 337eb20796bSJens Axboe * Lookup the (hopefully) full range of pages we need. 33882aa5d61SJens Axboe */ 33935f3d14dSJens Axboe spd.nr_pages = find_get_pages_contig(mapping, index, nr_pages, spd.pages); 340431a4820SFengguang Wu index += spd.nr_pages; 341eb20796bSJens Axboe 3425274f052SJens Axboe /* 343eb20796bSJens Axboe * If find_get_pages_contig() returned fewer pages than we needed, 344431a4820SFengguang Wu * readahead/allocate the rest and fill in the holes. 345eb20796bSJens Axboe */ 346431a4820SFengguang Wu if (spd.nr_pages < nr_pages) 347cf914a7dSRusty Russell page_cache_sync_readahead(mapping, &in->f_ra, in, 348cf914a7dSRusty Russell index, req_pages - spd.nr_pages); 349431a4820SFengguang Wu 350932cc6d4SJens Axboe error = 0; 351eb20796bSJens Axboe while (spd.nr_pages < nr_pages) { 352eb20796bSJens Axboe /* 353eb20796bSJens Axboe * Page could be there, find_get_pages_contig() breaks on 354eb20796bSJens Axboe * the first hole. 3555274f052SJens Axboe */ 3567480a904SJens Axboe page = find_get_page(mapping, index); 3577480a904SJens Axboe if (!page) { 358e27dedd8SJens Axboe /* 359eb20796bSJens Axboe * page didn't exist, allocate one. 3607480a904SJens Axboe */ 3617480a904SJens Axboe page = page_cache_alloc_cold(mapping); 3625274f052SJens Axboe if (!page) 3635274f052SJens Axboe break; 3645274f052SJens Axboe 3657480a904SJens Axboe error = add_to_page_cache_lru(page, mapping, index, 3660ae0b5d0SNick Piggin GFP_KERNEL); 3675274f052SJens Axboe if (unlikely(error)) { 3685274f052SJens Axboe page_cache_release(page); 369a0548871SJens Axboe if (error == -EEXIST) 370a0548871SJens Axboe continue; 3715274f052SJens Axboe break; 3725274f052SJens Axboe } 373eb20796bSJens Axboe /* 374eb20796bSJens Axboe * add_to_page_cache() locks the page, unlock it 375eb20796bSJens Axboe * to avoid convoluting the logic below even more. 376eb20796bSJens Axboe */ 377eb20796bSJens Axboe unlock_page(page); 3785274f052SJens Axboe } 3797480a904SJens Axboe 38035f3d14dSJens Axboe spd.pages[spd.nr_pages++] = page; 381eb20796bSJens Axboe index++; 382eb20796bSJens Axboe } 383eb20796bSJens Axboe 384eb20796bSJens Axboe /* 385eb20796bSJens Axboe * Now loop over the map and see if we need to start IO on any 386eb20796bSJens Axboe * pages, fill in the partial map, etc. 387eb20796bSJens Axboe */ 388eb20796bSJens Axboe index = *ppos >> PAGE_CACHE_SHIFT; 389eb20796bSJens Axboe nr_pages = spd.nr_pages; 390eb20796bSJens Axboe spd.nr_pages = 0; 391eb20796bSJens Axboe for (page_nr = 0; page_nr < nr_pages; page_nr++) { 392eb20796bSJens Axboe unsigned int this_len; 393eb20796bSJens Axboe 394eb20796bSJens Axboe if (!len) 395eb20796bSJens Axboe break; 396eb20796bSJens Axboe 397eb20796bSJens Axboe /* 398eb20796bSJens Axboe * this_len is the max we'll use from this page 399eb20796bSJens Axboe */ 400eb20796bSJens Axboe this_len = min_t(unsigned long, len, PAGE_CACHE_SIZE - loff); 40135f3d14dSJens Axboe page = spd.pages[page_nr]; 402eb20796bSJens Axboe 403a08a166fSFengguang Wu if (PageReadahead(page)) 404cf914a7dSRusty Russell page_cache_async_readahead(mapping, &in->f_ra, in, 405d8983910SFengguang Wu page, index, req_pages - page_nr); 406a08a166fSFengguang Wu 4077480a904SJens Axboe /* 4087480a904SJens Axboe * If the page isn't uptodate, we may need to start io on it 4097480a904SJens Axboe */ 4107480a904SJens Axboe if (!PageUptodate(page)) { 4117480a904SJens Axboe lock_page(page); 4127480a904SJens Axboe 4137480a904SJens Axboe /* 41432502b84SMiklos Szeredi * Page was truncated, or invalidated by the 41532502b84SMiklos Szeredi * filesystem. Redo the find/create, but this time the 41632502b84SMiklos Szeredi * page is kept locked, so there's no chance of another 41732502b84SMiklos Szeredi * race with truncate/invalidate. 4187480a904SJens Axboe */ 4197480a904SJens Axboe if (!page->mapping) { 4207480a904SJens Axboe unlock_page(page); 42132502b84SMiklos Szeredi page = find_or_create_page(mapping, index, 42232502b84SMiklos Szeredi mapping_gfp_mask(mapping)); 42332502b84SMiklos Szeredi 42432502b84SMiklos Szeredi if (!page) { 42532502b84SMiklos Szeredi error = -ENOMEM; 4267480a904SJens Axboe break; 4277480a904SJens Axboe } 42835f3d14dSJens Axboe page_cache_release(spd.pages[page_nr]); 42935f3d14dSJens Axboe spd.pages[page_nr] = page; 43032502b84SMiklos Szeredi } 4317480a904SJens Axboe /* 4327480a904SJens Axboe * page was already under io and is now done, great 4337480a904SJens Axboe */ 4347480a904SJens Axboe if (PageUptodate(page)) { 4357480a904SJens Axboe unlock_page(page); 4367480a904SJens Axboe goto fill_it; 4377480a904SJens Axboe } 4387480a904SJens Axboe 4397480a904SJens Axboe /* 4407480a904SJens Axboe * need to read in the page 4417480a904SJens Axboe */ 4427480a904SJens Axboe error = mapping->a_ops->readpage(in, page); 4437480a904SJens Axboe if (unlikely(error)) { 444eb20796bSJens Axboe /* 445eb20796bSJens Axboe * We really should re-lookup the page here, 446eb20796bSJens Axboe * but it complicates things a lot. Instead 447eb20796bSJens Axboe * lets just do what we already stored, and 448eb20796bSJens Axboe * we'll get it the next time we are called. 449eb20796bSJens Axboe */ 4507480a904SJens Axboe if (error == AOP_TRUNCATED_PAGE) 451eb20796bSJens Axboe error = 0; 452eb20796bSJens Axboe 4537480a904SJens Axboe break; 4547480a904SJens Axboe } 455620a324bSJens Axboe } 456620a324bSJens Axboe fill_it: 45791ad66efSJens Axboe /* 458620a324bSJens Axboe * i_size must be checked after PageUptodate. 45991ad66efSJens Axboe */ 46091ad66efSJens Axboe isize = i_size_read(mapping->host); 46191ad66efSJens Axboe end_index = (isize - 1) >> PAGE_CACHE_SHIFT; 462eb20796bSJens Axboe if (unlikely(!isize || index > end_index)) 46391ad66efSJens Axboe break; 46491ad66efSJens Axboe 46591ad66efSJens Axboe /* 46691ad66efSJens Axboe * if this is the last page, see if we need to shrink 46791ad66efSJens Axboe * the length and stop 46891ad66efSJens Axboe */ 46991ad66efSJens Axboe if (end_index == index) { 470475ecadeSHugh Dickins unsigned int plen; 471475ecadeSHugh Dickins 472475ecadeSHugh Dickins /* 473475ecadeSHugh Dickins * max good bytes in this page 474475ecadeSHugh Dickins */ 475475ecadeSHugh Dickins plen = ((isize - 1) & ~PAGE_CACHE_MASK) + 1; 476475ecadeSHugh Dickins if (plen <= loff) 47791ad66efSJens Axboe break; 478475ecadeSHugh Dickins 47991ad66efSJens Axboe /* 48091ad66efSJens Axboe * force quit after adding this page 48191ad66efSJens Axboe */ 482475ecadeSHugh Dickins this_len = min(this_len, plen - loff); 483eb20796bSJens Axboe len = this_len; 48491ad66efSJens Axboe } 485620a324bSJens Axboe 48635f3d14dSJens Axboe spd.partial[page_nr].offset = loff; 48735f3d14dSJens Axboe spd.partial[page_nr].len = this_len; 48882aa5d61SJens Axboe len -= this_len; 48991ad66efSJens Axboe loff = 0; 490eb20796bSJens Axboe spd.nr_pages++; 491eb20796bSJens Axboe index++; 4925274f052SJens Axboe } 4935274f052SJens Axboe 494eb20796bSJens Axboe /* 495475ecadeSHugh Dickins * Release any pages at the end, if we quit early. 'page_nr' is how far 496eb20796bSJens Axboe * we got, 'nr_pages' is how many pages are in the map. 497eb20796bSJens Axboe */ 498eb20796bSJens Axboe while (page_nr < nr_pages) 49935f3d14dSJens Axboe page_cache_release(spd.pages[page_nr++]); 500f4e6b498SFengguang Wu in->f_ra.prev_pos = (loff_t)index << PAGE_CACHE_SHIFT; 501eb20796bSJens Axboe 502912d35f8SJens Axboe if (spd.nr_pages) 50335f3d14dSJens Axboe error = splice_to_pipe(pipe, &spd); 50416c523ddSJens Axboe 505047fe360SEric Dumazet splice_shrink_spd(&spd); 5067480a904SJens Axboe return error; 5075274f052SJens Axboe } 5085274f052SJens Axboe 50983f9135bSJens Axboe /** 51083f9135bSJens Axboe * generic_file_splice_read - splice data from file to a pipe 51183f9135bSJens Axboe * @in: file to splice from 512932cc6d4SJens Axboe * @ppos: position in @in 51383f9135bSJens Axboe * @pipe: pipe to splice to 51483f9135bSJens Axboe * @len: number of bytes to splice 51583f9135bSJens Axboe * @flags: splice modifier flags 51683f9135bSJens Axboe * 517932cc6d4SJens Axboe * Description: 518932cc6d4SJens Axboe * Will read pages from given file and fill them into a pipe. Can be 519932cc6d4SJens Axboe * used as long as the address_space operations for the source implements 520932cc6d4SJens Axboe * a readpage() hook. 521932cc6d4SJens Axboe * 52283f9135bSJens Axboe */ 523cbb7e577SJens Axboe ssize_t generic_file_splice_read(struct file *in, loff_t *ppos, 524cbb7e577SJens Axboe struct pipe_inode_info *pipe, size_t len, 525cbb7e577SJens Axboe unsigned int flags) 5265274f052SJens Axboe { 527d366d398SJens Axboe loff_t isize, left; 5288191ecd1SJens Axboe int ret; 529d366d398SJens Axboe 530d366d398SJens Axboe isize = i_size_read(in->f_mapping->host); 531d366d398SJens Axboe if (unlikely(*ppos >= isize)) 532d366d398SJens Axboe return 0; 533d366d398SJens Axboe 534d366d398SJens Axboe left = isize - *ppos; 535d366d398SJens Axboe if (unlikely(left < len)) 536d366d398SJens Axboe len = left; 5375274f052SJens Axboe 538cbb7e577SJens Axboe ret = __generic_file_splice_read(in, ppos, pipe, len, flags); 539723590edSMiklos Szeredi if (ret > 0) { 540cbb7e577SJens Axboe *ppos += ret; 541723590edSMiklos Szeredi file_accessed(in); 542723590edSMiklos Szeredi } 5435274f052SJens Axboe 5445274f052SJens Axboe return ret; 5455274f052SJens Axboe } 546059a8f37SJens Axboe EXPORT_SYMBOL(generic_file_splice_read); 547059a8f37SJens Axboe 5486818173bSMiklos Szeredi static const struct pipe_buf_operations default_pipe_buf_ops = { 5496818173bSMiklos Szeredi .can_merge = 0, 5506818173bSMiklos Szeredi .map = generic_pipe_buf_map, 5516818173bSMiklos Szeredi .unmap = generic_pipe_buf_unmap, 5526818173bSMiklos Szeredi .confirm = generic_pipe_buf_confirm, 5536818173bSMiklos Szeredi .release = generic_pipe_buf_release, 5546818173bSMiklos Szeredi .steal = generic_pipe_buf_steal, 5556818173bSMiklos Szeredi .get = generic_pipe_buf_get, 5566818173bSMiklos Szeredi }; 5576818173bSMiklos Szeredi 5586818173bSMiklos Szeredi static ssize_t kernel_readv(struct file *file, const struct iovec *vec, 5596818173bSMiklos Szeredi unsigned long vlen, loff_t offset) 5606818173bSMiklos Szeredi { 5616818173bSMiklos Szeredi mm_segment_t old_fs; 5626818173bSMiklos Szeredi loff_t pos = offset; 5636818173bSMiklos Szeredi ssize_t res; 5646818173bSMiklos Szeredi 5656818173bSMiklos Szeredi old_fs = get_fs(); 5666818173bSMiklos Szeredi set_fs(get_ds()); 5676818173bSMiklos Szeredi /* The cast to a user pointer is valid due to the set_fs() */ 5686818173bSMiklos Szeredi res = vfs_readv(file, (const struct iovec __user *)vec, vlen, &pos); 5696818173bSMiklos Szeredi set_fs(old_fs); 5706818173bSMiklos Szeredi 5716818173bSMiklos Szeredi return res; 5726818173bSMiklos Szeredi } 5736818173bSMiklos Szeredi 5747bb307e8SAl Viro ssize_t kernel_write(struct file *file, const char *buf, size_t count, 575b2858d7dSMiklos Szeredi loff_t pos) 5760b0a47f5SMiklos Szeredi { 5770b0a47f5SMiklos Szeredi mm_segment_t old_fs; 5780b0a47f5SMiklos Szeredi ssize_t res; 5790b0a47f5SMiklos Szeredi 5800b0a47f5SMiklos Szeredi old_fs = get_fs(); 5810b0a47f5SMiklos Szeredi set_fs(get_ds()); 5820b0a47f5SMiklos Szeredi /* The cast to a user pointer is valid due to the set_fs() */ 5837bb307e8SAl Viro res = vfs_write(file, (__force const char __user *)buf, count, &pos); 5840b0a47f5SMiklos Szeredi set_fs(old_fs); 5850b0a47f5SMiklos Szeredi 5860b0a47f5SMiklos Szeredi return res; 5870b0a47f5SMiklos Szeredi } 5887bb307e8SAl Viro EXPORT_SYMBOL(kernel_write); 5890b0a47f5SMiklos Szeredi 5906818173bSMiklos Szeredi ssize_t default_file_splice_read(struct file *in, loff_t *ppos, 5916818173bSMiklos Szeredi struct pipe_inode_info *pipe, size_t len, 5926818173bSMiklos Szeredi unsigned int flags) 5936818173bSMiklos Szeredi { 5946818173bSMiklos Szeredi unsigned int nr_pages; 5956818173bSMiklos Szeredi unsigned int nr_freed; 5966818173bSMiklos Szeredi size_t offset; 59735f3d14dSJens Axboe struct page *pages[PIPE_DEF_BUFFERS]; 59835f3d14dSJens Axboe struct partial_page partial[PIPE_DEF_BUFFERS]; 59935f3d14dSJens Axboe struct iovec *vec, __vec[PIPE_DEF_BUFFERS]; 6006818173bSMiklos Szeredi ssize_t res; 6016818173bSMiklos Szeredi size_t this_len; 6026818173bSMiklos Szeredi int error; 6036818173bSMiklos Szeredi int i; 6046818173bSMiklos Szeredi struct splice_pipe_desc spd = { 6056818173bSMiklos Szeredi .pages = pages, 6066818173bSMiklos Szeredi .partial = partial, 607047fe360SEric Dumazet .nr_pages_max = PIPE_DEF_BUFFERS, 6086818173bSMiklos Szeredi .flags = flags, 6096818173bSMiklos Szeredi .ops = &default_pipe_buf_ops, 6106818173bSMiklos Szeredi .spd_release = spd_release_page, 6116818173bSMiklos Szeredi }; 6126818173bSMiklos Szeredi 61335f3d14dSJens Axboe if (splice_grow_spd(pipe, &spd)) 61435f3d14dSJens Axboe return -ENOMEM; 61535f3d14dSJens Axboe 61635f3d14dSJens Axboe res = -ENOMEM; 61735f3d14dSJens Axboe vec = __vec; 618047fe360SEric Dumazet if (spd.nr_pages_max > PIPE_DEF_BUFFERS) { 619047fe360SEric Dumazet vec = kmalloc(spd.nr_pages_max * sizeof(struct iovec), GFP_KERNEL); 62035f3d14dSJens Axboe if (!vec) 62135f3d14dSJens Axboe goto shrink_ret; 62235f3d14dSJens Axboe } 62335f3d14dSJens Axboe 6246818173bSMiklos Szeredi offset = *ppos & ~PAGE_CACHE_MASK; 6256818173bSMiklos Szeredi nr_pages = (len + offset + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; 6266818173bSMiklos Szeredi 627047fe360SEric Dumazet for (i = 0; i < nr_pages && i < spd.nr_pages_max && len; i++) { 6286818173bSMiklos Szeredi struct page *page; 6296818173bSMiklos Szeredi 6304f231228SJens Axboe page = alloc_page(GFP_USER); 6316818173bSMiklos Szeredi error = -ENOMEM; 6326818173bSMiklos Szeredi if (!page) 6336818173bSMiklos Szeredi goto err; 6346818173bSMiklos Szeredi 6356818173bSMiklos Szeredi this_len = min_t(size_t, len, PAGE_CACHE_SIZE - offset); 6364f231228SJens Axboe vec[i].iov_base = (void __user *) page_address(page); 6376818173bSMiklos Szeredi vec[i].iov_len = this_len; 63835f3d14dSJens Axboe spd.pages[i] = page; 6396818173bSMiklos Szeredi spd.nr_pages++; 6406818173bSMiklos Szeredi len -= this_len; 6416818173bSMiklos Szeredi offset = 0; 6426818173bSMiklos Szeredi } 6436818173bSMiklos Szeredi 6446818173bSMiklos Szeredi res = kernel_readv(in, vec, spd.nr_pages, *ppos); 64577f6bf57SAndrew Morton if (res < 0) { 64677f6bf57SAndrew Morton error = res; 6476818173bSMiklos Szeredi goto err; 64877f6bf57SAndrew Morton } 6496818173bSMiklos Szeredi 6506818173bSMiklos Szeredi error = 0; 6516818173bSMiklos Szeredi if (!res) 6526818173bSMiklos Szeredi goto err; 6536818173bSMiklos Szeredi 6546818173bSMiklos Szeredi nr_freed = 0; 6556818173bSMiklos Szeredi for (i = 0; i < spd.nr_pages; i++) { 6566818173bSMiklos Szeredi this_len = min_t(size_t, vec[i].iov_len, res); 65735f3d14dSJens Axboe spd.partial[i].offset = 0; 65835f3d14dSJens Axboe spd.partial[i].len = this_len; 6596818173bSMiklos Szeredi if (!this_len) { 66035f3d14dSJens Axboe __free_page(spd.pages[i]); 66135f3d14dSJens Axboe spd.pages[i] = NULL; 6626818173bSMiklos Szeredi nr_freed++; 6636818173bSMiklos Szeredi } 6646818173bSMiklos Szeredi res -= this_len; 6656818173bSMiklos Szeredi } 6666818173bSMiklos Szeredi spd.nr_pages -= nr_freed; 6676818173bSMiklos Szeredi 6686818173bSMiklos Szeredi res = splice_to_pipe(pipe, &spd); 6696818173bSMiklos Szeredi if (res > 0) 6706818173bSMiklos Szeredi *ppos += res; 6716818173bSMiklos Szeredi 67235f3d14dSJens Axboe shrink_ret: 67335f3d14dSJens Axboe if (vec != __vec) 67435f3d14dSJens Axboe kfree(vec); 675047fe360SEric Dumazet splice_shrink_spd(&spd); 6766818173bSMiklos Szeredi return res; 6776818173bSMiklos Szeredi 6786818173bSMiklos Szeredi err: 6794f231228SJens Axboe for (i = 0; i < spd.nr_pages; i++) 68035f3d14dSJens Axboe __free_page(spd.pages[i]); 6814f231228SJens Axboe 68235f3d14dSJens Axboe res = error; 68335f3d14dSJens Axboe goto shrink_ret; 6846818173bSMiklos Szeredi } 6856818173bSMiklos Szeredi EXPORT_SYMBOL(default_file_splice_read); 6866818173bSMiklos Szeredi 6875274f052SJens Axboe /* 6884f6f0bd2SJens Axboe * Send 'sd->len' bytes to socket from 'sd->file' at position 'sd->pos' 689016b661eSJens Axboe * using sendpage(). Return the number of bytes sent. 6905274f052SJens Axboe */ 69176ad4d11SJens Axboe static int pipe_to_sendpage(struct pipe_inode_info *pipe, 6925274f052SJens Axboe struct pipe_buffer *buf, struct splice_desc *sd) 6935274f052SJens Axboe { 6946a14b90bSJens Axboe struct file *file = sd->u.file; 6955274f052SJens Axboe loff_t pos = sd->pos; 696a8adbe37SMichał Mirosław int more; 6975274f052SJens Axboe 698a8adbe37SMichał Mirosław if (!likely(file->f_op && file->f_op->sendpage)) 699a8adbe37SMichał Mirosław return -EINVAL; 700a8adbe37SMichał Mirosław 70135f9c09fSEric Dumazet more = (sd->flags & SPLICE_F_MORE) ? MSG_MORE : 0; 702ae62ca7bSEric Dumazet 703ae62ca7bSEric Dumazet if (sd->len < sd->total_len && pipe->nrbufs > 1) 70435f9c09fSEric Dumazet more |= MSG_SENDPAGE_NOTLAST; 705ae62ca7bSEric Dumazet 706a8adbe37SMichał Mirosław return file->f_op->sendpage(file, buf->page, buf->offset, 707f84d7519SJens Axboe sd->len, &pos, more); 7085274f052SJens Axboe } 7095274f052SJens Axboe 7105274f052SJens Axboe /* 7115274f052SJens Axboe * This is a little more tricky than the file -> pipe splicing. There are 7125274f052SJens Axboe * basically three cases: 7135274f052SJens Axboe * 7145274f052SJens Axboe * - Destination page already exists in the address space and there 7155274f052SJens Axboe * are users of it. For that case we have no other option that 7165274f052SJens Axboe * copying the data. Tough luck. 7175274f052SJens Axboe * - Destination page already exists in the address space, but there 7185274f052SJens Axboe * are no users of it. Make sure it's uptodate, then drop it. Fall 7195274f052SJens Axboe * through to last case. 7205274f052SJens Axboe * - Destination page does not exist, we can add the pipe page to 7215274f052SJens Axboe * the page cache and avoid the copy. 7225274f052SJens Axboe * 72383f9135bSJens Axboe * If asked to move pages to the output file (SPLICE_F_MOVE is set in 72483f9135bSJens Axboe * sd->flags), we attempt to migrate pages from the pipe to the output 72583f9135bSJens Axboe * file address space page cache. This is possible if no one else has 72683f9135bSJens Axboe * the pipe page referenced outside of the pipe and page cache. If 72783f9135bSJens Axboe * SPLICE_F_MOVE isn't set, or we cannot move the page, we simply create 72883f9135bSJens Axboe * a new page in the output file page cache and fill/dirty that. 7295274f052SJens Axboe */ 730328eaabaSMiklos Szeredi int pipe_to_file(struct pipe_inode_info *pipe, struct pipe_buffer *buf, 7315274f052SJens Axboe struct splice_desc *sd) 7325274f052SJens Axboe { 7336a14b90bSJens Axboe struct file *file = sd->u.file; 7345274f052SJens Axboe struct address_space *mapping = file->f_mapping; 735016b661eSJens Axboe unsigned int offset, this_len; 7365274f052SJens Axboe struct page *page; 737afddba49SNick Piggin void *fsdata; 7383e7ee3e7SJens Axboe int ret; 7395274f052SJens Axboe 7405274f052SJens Axboe offset = sd->pos & ~PAGE_CACHE_MASK; 7415274f052SJens Axboe 742016b661eSJens Axboe this_len = sd->len; 743016b661eSJens Axboe if (this_len + offset > PAGE_CACHE_SIZE) 744016b661eSJens Axboe this_len = PAGE_CACHE_SIZE - offset; 745016b661eSJens Axboe 746afddba49SNick Piggin ret = pagecache_write_begin(file, mapping, sd->pos, this_len, 747afddba49SNick Piggin AOP_FLAG_UNINTERRUPTIBLE, &page, &fsdata); 7489e0267c2SJens Axboe if (unlikely(ret)) 749afddba49SNick Piggin goto out; 7505274f052SJens Axboe 7510568b409SJens Axboe if (buf->page != page) { 75276ad4d11SJens Axboe char *src = buf->ops->map(pipe, buf, 1); 753e8e3c3d6SCong Wang char *dst = kmap_atomic(page); 7545abc97aaSJens Axboe 755016b661eSJens Axboe memcpy(dst + offset, src + buf->offset, this_len); 7565274f052SJens Axboe flush_dcache_page(page); 757e8e3c3d6SCong Wang kunmap_atomic(dst); 75876ad4d11SJens Axboe buf->ops->unmap(pipe, buf, src); 7595abc97aaSJens Axboe } 760afddba49SNick Piggin ret = pagecache_write_end(file, mapping, sd->pos, this_len, this_len, 761afddba49SNick Piggin page, fsdata); 7620568b409SJens Axboe out: 7635274f052SJens Axboe return ret; 7645274f052SJens Axboe } 765328eaabaSMiklos Szeredi EXPORT_SYMBOL(pipe_to_file); 7665274f052SJens Axboe 767b3c2d2ddSMiklos Szeredi static void wakeup_pipe_writers(struct pipe_inode_info *pipe) 768b3c2d2ddSMiklos Szeredi { 769b3c2d2ddSMiklos Szeredi smp_mb(); 770b3c2d2ddSMiklos Szeredi if (waitqueue_active(&pipe->wait)) 771b3c2d2ddSMiklos Szeredi wake_up_interruptible(&pipe->wait); 772b3c2d2ddSMiklos Szeredi kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT); 773b3c2d2ddSMiklos Szeredi } 774b3c2d2ddSMiklos Szeredi 775b3c2d2ddSMiklos Szeredi /** 776b3c2d2ddSMiklos Szeredi * splice_from_pipe_feed - feed available data from a pipe to a file 777b3c2d2ddSMiklos Szeredi * @pipe: pipe to splice from 778b3c2d2ddSMiklos Szeredi * @sd: information to @actor 779b3c2d2ddSMiklos Szeredi * @actor: handler that splices the data 780b3c2d2ddSMiklos Szeredi * 781b3c2d2ddSMiklos Szeredi * Description: 782b3c2d2ddSMiklos Szeredi * This function loops over the pipe and calls @actor to do the 783b3c2d2ddSMiklos Szeredi * actual moving of a single struct pipe_buffer to the desired 784b3c2d2ddSMiklos Szeredi * destination. It returns when there's no more buffers left in 785b3c2d2ddSMiklos Szeredi * the pipe or if the requested number of bytes (@sd->total_len) 786b3c2d2ddSMiklos Szeredi * have been copied. It returns a positive number (one) if the 787b3c2d2ddSMiklos Szeredi * pipe needs to be filled with more data, zero if the required 788b3c2d2ddSMiklos Szeredi * number of bytes have been copied and -errno on error. 789b3c2d2ddSMiklos Szeredi * 790b3c2d2ddSMiklos Szeredi * This, together with splice_from_pipe_{begin,end,next}, may be 791b3c2d2ddSMiklos Szeredi * used to implement the functionality of __splice_from_pipe() when 792b3c2d2ddSMiklos Szeredi * locking is required around copying the pipe buffers to the 793b3c2d2ddSMiklos Szeredi * destination. 794b3c2d2ddSMiklos Szeredi */ 795b3c2d2ddSMiklos Szeredi int splice_from_pipe_feed(struct pipe_inode_info *pipe, struct splice_desc *sd, 796b3c2d2ddSMiklos Szeredi splice_actor *actor) 797b3c2d2ddSMiklos Szeredi { 798b3c2d2ddSMiklos Szeredi int ret; 799b3c2d2ddSMiklos Szeredi 800b3c2d2ddSMiklos Szeredi while (pipe->nrbufs) { 801b3c2d2ddSMiklos Szeredi struct pipe_buffer *buf = pipe->bufs + pipe->curbuf; 802b3c2d2ddSMiklos Szeredi const struct pipe_buf_operations *ops = buf->ops; 803b3c2d2ddSMiklos Szeredi 804b3c2d2ddSMiklos Szeredi sd->len = buf->len; 805b3c2d2ddSMiklos Szeredi if (sd->len > sd->total_len) 806b3c2d2ddSMiklos Szeredi sd->len = sd->total_len; 807b3c2d2ddSMiklos Szeredi 808a8adbe37SMichał Mirosław ret = buf->ops->confirm(pipe, buf); 809a8adbe37SMichał Mirosław if (unlikely(ret)) { 810b3c2d2ddSMiklos Szeredi if (ret == -ENODATA) 811b3c2d2ddSMiklos Szeredi ret = 0; 812b3c2d2ddSMiklos Szeredi return ret; 813b3c2d2ddSMiklos Szeredi } 814a8adbe37SMichał Mirosław 815a8adbe37SMichał Mirosław ret = actor(pipe, buf, sd); 816a8adbe37SMichał Mirosław if (ret <= 0) 817a8adbe37SMichał Mirosław return ret; 818a8adbe37SMichał Mirosław 819b3c2d2ddSMiklos Szeredi buf->offset += ret; 820b3c2d2ddSMiklos Szeredi buf->len -= ret; 821b3c2d2ddSMiklos Szeredi 822b3c2d2ddSMiklos Szeredi sd->num_spliced += ret; 823b3c2d2ddSMiklos Szeredi sd->len -= ret; 824b3c2d2ddSMiklos Szeredi sd->pos += ret; 825b3c2d2ddSMiklos Szeredi sd->total_len -= ret; 826b3c2d2ddSMiklos Szeredi 827b3c2d2ddSMiklos Szeredi if (!buf->len) { 828b3c2d2ddSMiklos Szeredi buf->ops = NULL; 829b3c2d2ddSMiklos Szeredi ops->release(pipe, buf); 83035f3d14dSJens Axboe pipe->curbuf = (pipe->curbuf + 1) & (pipe->buffers - 1); 831b3c2d2ddSMiklos Szeredi pipe->nrbufs--; 8326447a3cfSAl Viro if (pipe->files) 833b3c2d2ddSMiklos Szeredi sd->need_wakeup = true; 834b3c2d2ddSMiklos Szeredi } 835b3c2d2ddSMiklos Szeredi 836b3c2d2ddSMiklos Szeredi if (!sd->total_len) 837b3c2d2ddSMiklos Szeredi return 0; 838b3c2d2ddSMiklos Szeredi } 839b3c2d2ddSMiklos Szeredi 840b3c2d2ddSMiklos Szeredi return 1; 841b3c2d2ddSMiklos Szeredi } 842b3c2d2ddSMiklos Szeredi EXPORT_SYMBOL(splice_from_pipe_feed); 843b3c2d2ddSMiklos Szeredi 844b3c2d2ddSMiklos Szeredi /** 845b3c2d2ddSMiklos Szeredi * splice_from_pipe_next - wait for some data to splice from 846b3c2d2ddSMiklos Szeredi * @pipe: pipe to splice from 847b3c2d2ddSMiklos Szeredi * @sd: information about the splice operation 848b3c2d2ddSMiklos Szeredi * 849b3c2d2ddSMiklos Szeredi * Description: 850b3c2d2ddSMiklos Szeredi * This function will wait for some data and return a positive 851b3c2d2ddSMiklos Szeredi * value (one) if pipe buffers are available. It will return zero 852b3c2d2ddSMiklos Szeredi * or -errno if no more data needs to be spliced. 853b3c2d2ddSMiklos Szeredi */ 854b3c2d2ddSMiklos Szeredi int splice_from_pipe_next(struct pipe_inode_info *pipe, struct splice_desc *sd) 855b3c2d2ddSMiklos Szeredi { 856b3c2d2ddSMiklos Szeredi while (!pipe->nrbufs) { 857b3c2d2ddSMiklos Szeredi if (!pipe->writers) 858b3c2d2ddSMiklos Szeredi return 0; 859b3c2d2ddSMiklos Szeredi 860b3c2d2ddSMiklos Szeredi if (!pipe->waiting_writers && sd->num_spliced) 861b3c2d2ddSMiklos Szeredi return 0; 862b3c2d2ddSMiklos Szeredi 863b3c2d2ddSMiklos Szeredi if (sd->flags & SPLICE_F_NONBLOCK) 864b3c2d2ddSMiklos Szeredi return -EAGAIN; 865b3c2d2ddSMiklos Szeredi 866b3c2d2ddSMiklos Szeredi if (signal_pending(current)) 867b3c2d2ddSMiklos Szeredi return -ERESTARTSYS; 868b3c2d2ddSMiklos Szeredi 869b3c2d2ddSMiklos Szeredi if (sd->need_wakeup) { 870b3c2d2ddSMiklos Szeredi wakeup_pipe_writers(pipe); 871b3c2d2ddSMiklos Szeredi sd->need_wakeup = false; 872b3c2d2ddSMiklos Szeredi } 873b3c2d2ddSMiklos Szeredi 874b3c2d2ddSMiklos Szeredi pipe_wait(pipe); 875b3c2d2ddSMiklos Szeredi } 876b3c2d2ddSMiklos Szeredi 877b3c2d2ddSMiklos Szeredi return 1; 878b3c2d2ddSMiklos Szeredi } 879b3c2d2ddSMiklos Szeredi EXPORT_SYMBOL(splice_from_pipe_next); 880b3c2d2ddSMiklos Szeredi 881b3c2d2ddSMiklos Szeredi /** 882b3c2d2ddSMiklos Szeredi * splice_from_pipe_begin - start splicing from pipe 883b80901bbSRandy Dunlap * @sd: information about the splice operation 884b3c2d2ddSMiklos Szeredi * 885b3c2d2ddSMiklos Szeredi * Description: 886b3c2d2ddSMiklos Szeredi * This function should be called before a loop containing 887b3c2d2ddSMiklos Szeredi * splice_from_pipe_next() and splice_from_pipe_feed() to 888b3c2d2ddSMiklos Szeredi * initialize the necessary fields of @sd. 889b3c2d2ddSMiklos Szeredi */ 890b3c2d2ddSMiklos Szeredi void splice_from_pipe_begin(struct splice_desc *sd) 891b3c2d2ddSMiklos Szeredi { 892b3c2d2ddSMiklos Szeredi sd->num_spliced = 0; 893b3c2d2ddSMiklos Szeredi sd->need_wakeup = false; 894b3c2d2ddSMiklos Szeredi } 895b3c2d2ddSMiklos Szeredi EXPORT_SYMBOL(splice_from_pipe_begin); 896b3c2d2ddSMiklos Szeredi 897b3c2d2ddSMiklos Szeredi /** 898b3c2d2ddSMiklos Szeredi * splice_from_pipe_end - finish splicing from pipe 899b3c2d2ddSMiklos Szeredi * @pipe: pipe to splice from 900b3c2d2ddSMiklos Szeredi * @sd: information about the splice operation 901b3c2d2ddSMiklos Szeredi * 902b3c2d2ddSMiklos Szeredi * Description: 903b3c2d2ddSMiklos Szeredi * This function will wake up pipe writers if necessary. It should 904b3c2d2ddSMiklos Szeredi * be called after a loop containing splice_from_pipe_next() and 905b3c2d2ddSMiklos Szeredi * splice_from_pipe_feed(). 906b3c2d2ddSMiklos Szeredi */ 907b3c2d2ddSMiklos Szeredi void splice_from_pipe_end(struct pipe_inode_info *pipe, struct splice_desc *sd) 908b3c2d2ddSMiklos Szeredi { 909b3c2d2ddSMiklos Szeredi if (sd->need_wakeup) 910b3c2d2ddSMiklos Szeredi wakeup_pipe_writers(pipe); 911b3c2d2ddSMiklos Szeredi } 912b3c2d2ddSMiklos Szeredi EXPORT_SYMBOL(splice_from_pipe_end); 913b3c2d2ddSMiklos Szeredi 914932cc6d4SJens Axboe /** 915932cc6d4SJens Axboe * __splice_from_pipe - splice data from a pipe to given actor 916932cc6d4SJens Axboe * @pipe: pipe to splice from 917932cc6d4SJens Axboe * @sd: information to @actor 918932cc6d4SJens Axboe * @actor: handler that splices the data 919932cc6d4SJens Axboe * 920932cc6d4SJens Axboe * Description: 921932cc6d4SJens Axboe * This function does little more than loop over the pipe and call 922932cc6d4SJens Axboe * @actor to do the actual moving of a single struct pipe_buffer to 923932cc6d4SJens Axboe * the desired destination. See pipe_to_file, pipe_to_sendpage, or 924932cc6d4SJens Axboe * pipe_to_user. 925932cc6d4SJens Axboe * 92683f9135bSJens Axboe */ 927c66ab6faSJens Axboe ssize_t __splice_from_pipe(struct pipe_inode_info *pipe, struct splice_desc *sd, 928c66ab6faSJens Axboe splice_actor *actor) 9295274f052SJens Axboe { 930b3c2d2ddSMiklos Szeredi int ret; 9315274f052SJens Axboe 932b3c2d2ddSMiklos Szeredi splice_from_pipe_begin(sd); 933b3c2d2ddSMiklos Szeredi do { 934b3c2d2ddSMiklos Szeredi ret = splice_from_pipe_next(pipe, sd); 935b3c2d2ddSMiklos Szeredi if (ret > 0) 936b3c2d2ddSMiklos Szeredi ret = splice_from_pipe_feed(pipe, sd, actor); 937b3c2d2ddSMiklos Szeredi } while (ret > 0); 938b3c2d2ddSMiklos Szeredi splice_from_pipe_end(pipe, sd); 9395274f052SJens Axboe 940b3c2d2ddSMiklos Szeredi return sd->num_spliced ? sd->num_spliced : ret; 9415274f052SJens Axboe } 94240bee44eSMark Fasheh EXPORT_SYMBOL(__splice_from_pipe); 9435274f052SJens Axboe 944932cc6d4SJens Axboe /** 945932cc6d4SJens Axboe * splice_from_pipe - splice data from a pipe to a file 946932cc6d4SJens Axboe * @pipe: pipe to splice from 947932cc6d4SJens Axboe * @out: file to splice to 948932cc6d4SJens Axboe * @ppos: position in @out 949932cc6d4SJens Axboe * @len: how many bytes to splice 950932cc6d4SJens Axboe * @flags: splice modifier flags 951932cc6d4SJens Axboe * @actor: handler that splices the data 952932cc6d4SJens Axboe * 953932cc6d4SJens Axboe * Description: 9542933970bSMiklos Szeredi * See __splice_from_pipe. This function locks the pipe inode, 955932cc6d4SJens Axboe * otherwise it's identical to __splice_from_pipe(). 956932cc6d4SJens Axboe * 957932cc6d4SJens Axboe */ 9586da61809SMark Fasheh ssize_t splice_from_pipe(struct pipe_inode_info *pipe, struct file *out, 9596da61809SMark Fasheh loff_t *ppos, size_t len, unsigned int flags, 9606da61809SMark Fasheh splice_actor *actor) 9616da61809SMark Fasheh { 9626da61809SMark Fasheh ssize_t ret; 963c66ab6faSJens Axboe struct splice_desc sd = { 964c66ab6faSJens Axboe .total_len = len, 965c66ab6faSJens Axboe .flags = flags, 966c66ab6faSJens Axboe .pos = *ppos, 9676a14b90bSJens Axboe .u.file = out, 968c66ab6faSJens Axboe }; 9696da61809SMark Fasheh 97061e0d47cSMiklos Szeredi pipe_lock(pipe); 971c66ab6faSJens Axboe ret = __splice_from_pipe(pipe, &sd, actor); 97261e0d47cSMiklos Szeredi pipe_unlock(pipe); 9736da61809SMark Fasheh 9746da61809SMark Fasheh return ret; 9756da61809SMark Fasheh } 9766da61809SMark Fasheh 9776da61809SMark Fasheh /** 97883f9135bSJens Axboe * generic_file_splice_write - splice data from a pipe to a file 9793a326a2cSIngo Molnar * @pipe: pipe info 98083f9135bSJens Axboe * @out: file to write to 981932cc6d4SJens Axboe * @ppos: position in @out 98283f9135bSJens Axboe * @len: number of bytes to splice 98383f9135bSJens Axboe * @flags: splice modifier flags 98483f9135bSJens Axboe * 985932cc6d4SJens Axboe * Description: 98683f9135bSJens Axboe * Will either move or copy pages (determined by @flags options) from 98783f9135bSJens Axboe * the given pipe inode to the given file. 98883f9135bSJens Axboe * 98983f9135bSJens Axboe */ 9903a326a2cSIngo Molnar ssize_t 9913a326a2cSIngo Molnar generic_file_splice_write(struct pipe_inode_info *pipe, struct file *out, 992cbb7e577SJens Axboe loff_t *ppos, size_t len, unsigned int flags) 9935274f052SJens Axboe { 9944f6f0bd2SJens Axboe struct address_space *mapping = out->f_mapping; 9958c34e2d6SJens Axboe struct inode *inode = mapping->host; 9967f3d4ee1SMiklos Szeredi struct splice_desc sd = { 9977f3d4ee1SMiklos Szeredi .total_len = len, 9987f3d4ee1SMiklos Szeredi .flags = flags, 9997f3d4ee1SMiklos Szeredi .pos = *ppos, 10007f3d4ee1SMiklos Szeredi .u.file = out, 10017f3d4ee1SMiklos Szeredi }; 10023a326a2cSIngo Molnar ssize_t ret; 10038c34e2d6SJens Axboe 100461e0d47cSMiklos Szeredi pipe_lock(pipe); 1005eb443e5aSMiklos Szeredi 1006eb443e5aSMiklos Szeredi splice_from_pipe_begin(&sd); 1007eb443e5aSMiklos Szeredi do { 1008eb443e5aSMiklos Szeredi ret = splice_from_pipe_next(pipe, &sd); 1009eb443e5aSMiklos Szeredi if (ret <= 0) 1010eb443e5aSMiklos Szeredi break; 1011eb443e5aSMiklos Szeredi 1012eb443e5aSMiklos Szeredi mutex_lock_nested(&inode->i_mutex, I_MUTEX_CHILD); 1013eb443e5aSMiklos Szeredi ret = file_remove_suid(out); 1014723590edSMiklos Szeredi if (!ret) { 1015c3b2da31SJosef Bacik ret = file_update_time(out); 1016c3b2da31SJosef Bacik if (!ret) 1017c3b2da31SJosef Bacik ret = splice_from_pipe_feed(pipe, &sd, 1018c3b2da31SJosef Bacik pipe_to_file); 1019723590edSMiklos Szeredi } 1020eb443e5aSMiklos Szeredi mutex_unlock(&inode->i_mutex); 1021eb443e5aSMiklos Szeredi } while (ret > 0); 1022eb443e5aSMiklos Szeredi splice_from_pipe_end(pipe, &sd); 1023eb443e5aSMiklos Szeredi 102461e0d47cSMiklos Szeredi pipe_unlock(pipe); 1025eb443e5aSMiklos Szeredi 1026eb443e5aSMiklos Szeredi if (sd.num_spliced) 1027eb443e5aSMiklos Szeredi ret = sd.num_spliced; 1028eb443e5aSMiklos Szeredi 1029a4514ebdSJens Axboe if (ret > 0) { 10307f3d4ee1SMiklos Szeredi int err; 10317f3d4ee1SMiklos Szeredi 1032148f948bSJan Kara err = generic_write_sync(out, *ppos, ret); 10334f6f0bd2SJens Axboe if (err) 10344f6f0bd2SJens Axboe ret = err; 1035148f948bSJan Kara else 1036148f948bSJan Kara *ppos += ret; 1037d0e1d66bSNamjae Jeon balance_dirty_pages_ratelimited(mapping); 1038a4514ebdSJens Axboe } 10394f6f0bd2SJens Axboe 10404f6f0bd2SJens Axboe return ret; 10415274f052SJens Axboe } 10425274f052SJens Axboe 1043059a8f37SJens Axboe EXPORT_SYMBOL(generic_file_splice_write); 1044059a8f37SJens Axboe 1045b2858d7dSMiklos Szeredi static int write_pipe_buf(struct pipe_inode_info *pipe, struct pipe_buffer *buf, 1046b2858d7dSMiklos Szeredi struct splice_desc *sd) 10470b0a47f5SMiklos Szeredi { 1048b2858d7dSMiklos Szeredi int ret; 1049b2858d7dSMiklos Szeredi void *data; 105006ae43f3SAl Viro loff_t tmp = sd->pos; 1051b2858d7dSMiklos Szeredi 1052b2858d7dSMiklos Szeredi data = buf->ops->map(pipe, buf, 0); 105306ae43f3SAl Viro ret = __kernel_write(sd->u.file, data + buf->offset, sd->len, &tmp); 1054b2858d7dSMiklos Szeredi buf->ops->unmap(pipe, buf, data); 1055b2858d7dSMiklos Szeredi 1056b2858d7dSMiklos Szeredi return ret; 10570b0a47f5SMiklos Szeredi } 10580b0a47f5SMiklos Szeredi 10590b0a47f5SMiklos Szeredi static ssize_t default_file_splice_write(struct pipe_inode_info *pipe, 10600b0a47f5SMiklos Szeredi struct file *out, loff_t *ppos, 10610b0a47f5SMiklos Szeredi size_t len, unsigned int flags) 10620b0a47f5SMiklos Szeredi { 1063b2858d7dSMiklos Szeredi ssize_t ret; 10640b0a47f5SMiklos Szeredi 1065b2858d7dSMiklos Szeredi ret = splice_from_pipe(pipe, out, ppos, len, flags, write_pipe_buf); 1066b2858d7dSMiklos Szeredi if (ret > 0) 1067b2858d7dSMiklos Szeredi *ppos += ret; 10680b0a47f5SMiklos Szeredi 1069b2858d7dSMiklos Szeredi return ret; 10700b0a47f5SMiklos Szeredi } 10710b0a47f5SMiklos Szeredi 107283f9135bSJens Axboe /** 107383f9135bSJens Axboe * generic_splice_sendpage - splice data from a pipe to a socket 1074932cc6d4SJens Axboe * @pipe: pipe to splice from 107583f9135bSJens Axboe * @out: socket to write to 1076932cc6d4SJens Axboe * @ppos: position in @out 107783f9135bSJens Axboe * @len: number of bytes to splice 107883f9135bSJens Axboe * @flags: splice modifier flags 107983f9135bSJens Axboe * 1080932cc6d4SJens Axboe * Description: 108183f9135bSJens Axboe * Will send @len bytes from the pipe to a network socket. No data copying 108283f9135bSJens Axboe * is involved. 108383f9135bSJens Axboe * 108483f9135bSJens Axboe */ 10853a326a2cSIngo Molnar ssize_t generic_splice_sendpage(struct pipe_inode_info *pipe, struct file *out, 1086cbb7e577SJens Axboe loff_t *ppos, size_t len, unsigned int flags) 10875274f052SJens Axboe { 108800522fb4SJens Axboe return splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_sendpage); 10895274f052SJens Axboe } 10905274f052SJens Axboe 1091059a8f37SJens Axboe EXPORT_SYMBOL(generic_splice_sendpage); 1092a0f06780SJeff Garzik 109383f9135bSJens Axboe /* 109483f9135bSJens Axboe * Attempt to initiate a splice from pipe to file. 109583f9135bSJens Axboe */ 10963a326a2cSIngo Molnar static long do_splice_from(struct pipe_inode_info *pipe, struct file *out, 1097cbb7e577SJens Axboe loff_t *ppos, size_t len, unsigned int flags) 10985274f052SJens Axboe { 10990b0a47f5SMiklos Szeredi ssize_t (*splice_write)(struct pipe_inode_info *, struct file *, 11000b0a47f5SMiklos Szeredi loff_t *, size_t, unsigned int); 11015274f052SJens Axboe int ret; 11025274f052SJens Axboe 110349570e9bSJens Axboe if (unlikely(!(out->f_mode & FMODE_WRITE))) 11045274f052SJens Axboe return -EBADF; 11055274f052SJens Axboe 1106efc968d4SLinus Torvalds if (unlikely(out->f_flags & O_APPEND)) 1107efc968d4SLinus Torvalds return -EINVAL; 1108efc968d4SLinus Torvalds 1109cbb7e577SJens Axboe ret = rw_verify_area(WRITE, out, ppos, len); 11105274f052SJens Axboe if (unlikely(ret < 0)) 11115274f052SJens Axboe return ret; 11125274f052SJens Axboe 1113cc56f7deSChangli Gao if (out->f_op && out->f_op->splice_write) 11140b0a47f5SMiklos Szeredi splice_write = out->f_op->splice_write; 1115cc56f7deSChangli Gao else 11160b0a47f5SMiklos Szeredi splice_write = default_file_splice_write; 11170b0a47f5SMiklos Szeredi 11182dd8c9adSAl Viro file_start_write(out); 11192dd8c9adSAl Viro ret = splice_write(pipe, out, ppos, len, flags); 11202dd8c9adSAl Viro file_end_write(out); 11212dd8c9adSAl Viro return ret; 11225274f052SJens Axboe } 11235274f052SJens Axboe 112483f9135bSJens Axboe /* 112583f9135bSJens Axboe * Attempt to initiate a splice from a file to a pipe. 112683f9135bSJens Axboe */ 1127cbb7e577SJens Axboe static long do_splice_to(struct file *in, loff_t *ppos, 1128cbb7e577SJens Axboe struct pipe_inode_info *pipe, size_t len, 1129cbb7e577SJens Axboe unsigned int flags) 11305274f052SJens Axboe { 11316818173bSMiklos Szeredi ssize_t (*splice_read)(struct file *, loff_t *, 11326818173bSMiklos Szeredi struct pipe_inode_info *, size_t, unsigned int); 11335274f052SJens Axboe int ret; 11345274f052SJens Axboe 113549570e9bSJens Axboe if (unlikely(!(in->f_mode & FMODE_READ))) 11365274f052SJens Axboe return -EBADF; 11375274f052SJens Axboe 1138cbb7e577SJens Axboe ret = rw_verify_area(READ, in, ppos, len); 11395274f052SJens Axboe if (unlikely(ret < 0)) 11405274f052SJens Axboe return ret; 11415274f052SJens Axboe 1142cc56f7deSChangli Gao if (in->f_op && in->f_op->splice_read) 11436818173bSMiklos Szeredi splice_read = in->f_op->splice_read; 1144cc56f7deSChangli Gao else 11456818173bSMiklos Szeredi splice_read = default_file_splice_read; 11466818173bSMiklos Szeredi 11476818173bSMiklos Szeredi return splice_read(in, ppos, pipe, len, flags); 11485274f052SJens Axboe } 11495274f052SJens Axboe 1150932cc6d4SJens Axboe /** 1151932cc6d4SJens Axboe * splice_direct_to_actor - splices data directly between two non-pipes 1152932cc6d4SJens Axboe * @in: file to splice from 1153932cc6d4SJens Axboe * @sd: actor information on where to splice to 1154932cc6d4SJens Axboe * @actor: handles the data splicing 1155932cc6d4SJens Axboe * 1156932cc6d4SJens Axboe * Description: 1157932cc6d4SJens Axboe * This is a special case helper to splice directly between two 1158932cc6d4SJens Axboe * points, without requiring an explicit pipe. Internally an allocated 1159932cc6d4SJens Axboe * pipe is cached in the process, and reused during the lifetime of 1160932cc6d4SJens Axboe * that process. 1161932cc6d4SJens Axboe * 1162c66ab6faSJens Axboe */ 1163c66ab6faSJens Axboe ssize_t splice_direct_to_actor(struct file *in, struct splice_desc *sd, 1164c66ab6faSJens Axboe splice_direct_actor *actor) 1165b92ce558SJens Axboe { 1166b92ce558SJens Axboe struct pipe_inode_info *pipe; 1167b92ce558SJens Axboe long ret, bytes; 1168b92ce558SJens Axboe umode_t i_mode; 1169c66ab6faSJens Axboe size_t len; 1170c66ab6faSJens Axboe int i, flags; 1171b92ce558SJens Axboe 1172b92ce558SJens Axboe /* 1173b92ce558SJens Axboe * We require the input being a regular file, as we don't want to 1174b92ce558SJens Axboe * randomly drop data for eg socket -> socket splicing. Use the 1175b92ce558SJens Axboe * piped splicing for that! 1176b92ce558SJens Axboe */ 1177496ad9aaSAl Viro i_mode = file_inode(in)->i_mode; 1178b92ce558SJens Axboe if (unlikely(!S_ISREG(i_mode) && !S_ISBLK(i_mode))) 1179b92ce558SJens Axboe return -EINVAL; 1180b92ce558SJens Axboe 1181b92ce558SJens Axboe /* 1182b92ce558SJens Axboe * neither in nor out is a pipe, setup an internal pipe attached to 1183b92ce558SJens Axboe * 'out' and transfer the wanted data from 'in' to 'out' through that 1184b92ce558SJens Axboe */ 1185b92ce558SJens Axboe pipe = current->splice_pipe; 118649570e9bSJens Axboe if (unlikely(!pipe)) { 11877bee130eSAl Viro pipe = alloc_pipe_info(); 1188b92ce558SJens Axboe if (!pipe) 1189b92ce558SJens Axboe return -ENOMEM; 1190b92ce558SJens Axboe 1191b92ce558SJens Axboe /* 1192b92ce558SJens Axboe * We don't have an immediate reader, but we'll read the stuff 119300522fb4SJens Axboe * out of the pipe right after the splice_to_pipe(). So set 1194b92ce558SJens Axboe * PIPE_READERS appropriately. 1195b92ce558SJens Axboe */ 1196b92ce558SJens Axboe pipe->readers = 1; 1197b92ce558SJens Axboe 1198b92ce558SJens Axboe current->splice_pipe = pipe; 1199b92ce558SJens Axboe } 1200b92ce558SJens Axboe 1201b92ce558SJens Axboe /* 120273d62d83SIngo Molnar * Do the splice. 1203b92ce558SJens Axboe */ 1204b92ce558SJens Axboe ret = 0; 1205b92ce558SJens Axboe bytes = 0; 1206c66ab6faSJens Axboe len = sd->total_len; 1207c66ab6faSJens Axboe flags = sd->flags; 1208c66ab6faSJens Axboe 1209c66ab6faSJens Axboe /* 1210c66ab6faSJens Axboe * Don't block on output, we have to drain the direct pipe. 1211c66ab6faSJens Axboe */ 1212c66ab6faSJens Axboe sd->flags &= ~SPLICE_F_NONBLOCK; 1213b92ce558SJens Axboe 1214b92ce558SJens Axboe while (len) { 121551a92c0fSJens Axboe size_t read_len; 1216a82c53a0STom Zanussi loff_t pos = sd->pos, prev_pos = pos; 1217b92ce558SJens Axboe 1218bcd4f3acSJens Axboe ret = do_splice_to(in, &pos, pipe, len, flags); 121951a92c0fSJens Axboe if (unlikely(ret <= 0)) 1220b92ce558SJens Axboe goto out_release; 1221b92ce558SJens Axboe 1222b92ce558SJens Axboe read_len = ret; 1223c66ab6faSJens Axboe sd->total_len = read_len; 1224b92ce558SJens Axboe 1225b92ce558SJens Axboe /* 1226b92ce558SJens Axboe * NOTE: nonblocking mode only applies to the input. We 1227b92ce558SJens Axboe * must not do the output in nonblocking mode as then we 1228b92ce558SJens Axboe * could get stuck data in the internal pipe: 1229b92ce558SJens Axboe */ 1230c66ab6faSJens Axboe ret = actor(pipe, sd); 1231a82c53a0STom Zanussi if (unlikely(ret <= 0)) { 1232a82c53a0STom Zanussi sd->pos = prev_pos; 1233b92ce558SJens Axboe goto out_release; 1234a82c53a0STom Zanussi } 1235b92ce558SJens Axboe 1236b92ce558SJens Axboe bytes += ret; 1237b92ce558SJens Axboe len -= ret; 1238bcd4f3acSJens Axboe sd->pos = pos; 1239b92ce558SJens Axboe 1240a82c53a0STom Zanussi if (ret < read_len) { 1241a82c53a0STom Zanussi sd->pos = prev_pos + ret; 124251a92c0fSJens Axboe goto out_release; 1243b92ce558SJens Axboe } 1244a82c53a0STom Zanussi } 1245b92ce558SJens Axboe 12469e97198dSJens Axboe done: 1247b92ce558SJens Axboe pipe->nrbufs = pipe->curbuf = 0; 12489e97198dSJens Axboe file_accessed(in); 1249b92ce558SJens Axboe return bytes; 1250b92ce558SJens Axboe 1251b92ce558SJens Axboe out_release: 1252b92ce558SJens Axboe /* 1253b92ce558SJens Axboe * If we did an incomplete transfer we must release 1254b92ce558SJens Axboe * the pipe buffers in question: 1255b92ce558SJens Axboe */ 125635f3d14dSJens Axboe for (i = 0; i < pipe->buffers; i++) { 1257b92ce558SJens Axboe struct pipe_buffer *buf = pipe->bufs + i; 1258b92ce558SJens Axboe 1259b92ce558SJens Axboe if (buf->ops) { 1260b92ce558SJens Axboe buf->ops->release(pipe, buf); 1261b92ce558SJens Axboe buf->ops = NULL; 1262b92ce558SJens Axboe } 1263b92ce558SJens Axboe } 1264b92ce558SJens Axboe 12659e97198dSJens Axboe if (!bytes) 12669e97198dSJens Axboe bytes = ret; 1267b92ce558SJens Axboe 12689e97198dSJens Axboe goto done; 1269c66ab6faSJens Axboe } 1270c66ab6faSJens Axboe EXPORT_SYMBOL(splice_direct_to_actor); 1271c66ab6faSJens Axboe 1272c66ab6faSJens Axboe static int direct_splice_actor(struct pipe_inode_info *pipe, 1273c66ab6faSJens Axboe struct splice_desc *sd) 1274c66ab6faSJens Axboe { 12756a14b90bSJens Axboe struct file *file = sd->u.file; 1276c66ab6faSJens Axboe 12777995bd28SAl Viro return do_splice_from(pipe, file, sd->opos, sd->total_len, 12782cb4b05eSChangli Gao sd->flags); 1279c66ab6faSJens Axboe } 1280c66ab6faSJens Axboe 1281932cc6d4SJens Axboe /** 1282932cc6d4SJens Axboe * do_splice_direct - splices data directly between two files 1283932cc6d4SJens Axboe * @in: file to splice from 1284932cc6d4SJens Axboe * @ppos: input file offset 1285932cc6d4SJens Axboe * @out: file to splice to 1286acdb37c3SRandy Dunlap * @opos: output file offset 1287932cc6d4SJens Axboe * @len: number of bytes to splice 1288932cc6d4SJens Axboe * @flags: splice modifier flags 1289932cc6d4SJens Axboe * 1290932cc6d4SJens Axboe * Description: 1291932cc6d4SJens Axboe * For use by do_sendfile(). splice can easily emulate sendfile, but 1292932cc6d4SJens Axboe * doing it in the application would incur an extra system call 1293932cc6d4SJens Axboe * (splice in + splice out, as compared to just sendfile()). So this helper 1294932cc6d4SJens Axboe * can splice directly through a process-private pipe. 1295932cc6d4SJens Axboe * 1296932cc6d4SJens Axboe */ 1297c66ab6faSJens Axboe long do_splice_direct(struct file *in, loff_t *ppos, struct file *out, 12987995bd28SAl Viro loff_t *opos, size_t len, unsigned int flags) 1299c66ab6faSJens Axboe { 1300c66ab6faSJens Axboe struct splice_desc sd = { 1301c66ab6faSJens Axboe .len = len, 1302c66ab6faSJens Axboe .total_len = len, 1303c66ab6faSJens Axboe .flags = flags, 1304c66ab6faSJens Axboe .pos = *ppos, 13056a14b90bSJens Axboe .u.file = out, 13067995bd28SAl Viro .opos = opos, 1307c66ab6faSJens Axboe }; 130851a92c0fSJens Axboe long ret; 1309c66ab6faSJens Axboe 1310c66ab6faSJens Axboe ret = splice_direct_to_actor(in, &sd, direct_splice_actor); 131151a92c0fSJens Axboe if (ret > 0) 1312a82c53a0STom Zanussi *ppos = sd.pos; 131351a92c0fSJens Axboe 1314c66ab6faSJens Axboe return ret; 1315b92ce558SJens Axboe } 1316b92ce558SJens Axboe 13177c77f0b3SMiklos Szeredi static int splice_pipe_to_pipe(struct pipe_inode_info *ipipe, 13187c77f0b3SMiklos Szeredi struct pipe_inode_info *opipe, 13197c77f0b3SMiklos Szeredi size_t len, unsigned int flags); 1320ddac0d39SJens Axboe 1321ddac0d39SJens Axboe /* 132283f9135bSJens Axboe * Determine where to splice to/from. 132383f9135bSJens Axboe */ 1324529565dcSIngo Molnar static long do_splice(struct file *in, loff_t __user *off_in, 1325529565dcSIngo Molnar struct file *out, loff_t __user *off_out, 1326529565dcSIngo Molnar size_t len, unsigned int flags) 13275274f052SJens Axboe { 13287c77f0b3SMiklos Szeredi struct pipe_inode_info *ipipe; 13297c77f0b3SMiklos Szeredi struct pipe_inode_info *opipe; 13307995bd28SAl Viro loff_t offset; 1331a4514ebdSJens Axboe long ret; 13325274f052SJens Axboe 133371993e62SLinus Torvalds ipipe = get_pipe_info(in); 133471993e62SLinus Torvalds opipe = get_pipe_info(out); 13357c77f0b3SMiklos Szeredi 13367c77f0b3SMiklos Szeredi if (ipipe && opipe) { 13377c77f0b3SMiklos Szeredi if (off_in || off_out) 13387c77f0b3SMiklos Szeredi return -ESPIPE; 13397c77f0b3SMiklos Szeredi 13407c77f0b3SMiklos Szeredi if (!(in->f_mode & FMODE_READ)) 13417c77f0b3SMiklos Szeredi return -EBADF; 13427c77f0b3SMiklos Szeredi 13437c77f0b3SMiklos Szeredi if (!(out->f_mode & FMODE_WRITE)) 13447c77f0b3SMiklos Szeredi return -EBADF; 13457c77f0b3SMiklos Szeredi 13467c77f0b3SMiklos Szeredi /* Splicing to self would be fun, but... */ 13477c77f0b3SMiklos Szeredi if (ipipe == opipe) 13487c77f0b3SMiklos Szeredi return -EINVAL; 13497c77f0b3SMiklos Szeredi 13507c77f0b3SMiklos Szeredi return splice_pipe_to_pipe(ipipe, opipe, len, flags); 13517c77f0b3SMiklos Szeredi } 13527c77f0b3SMiklos Szeredi 13537c77f0b3SMiklos Szeredi if (ipipe) { 1354529565dcSIngo Molnar if (off_in) 1355529565dcSIngo Molnar return -ESPIPE; 1356b92ce558SJens Axboe if (off_out) { 135719c9a49bSChangli Gao if (!(out->f_mode & FMODE_PWRITE)) 1358b92ce558SJens Axboe return -EINVAL; 1359cbb7e577SJens Axboe if (copy_from_user(&offset, off_out, sizeof(loff_t))) 1360b92ce558SJens Axboe return -EFAULT; 13617995bd28SAl Viro } else { 13627995bd28SAl Viro offset = out->f_pos; 13637995bd28SAl Viro } 1364529565dcSIngo Molnar 13657995bd28SAl Viro ret = do_splice_from(ipipe, out, &offset, len, flags); 1366a4514ebdSJens Axboe 13677995bd28SAl Viro if (!off_out) 13687995bd28SAl Viro out->f_pos = offset; 13697995bd28SAl Viro else if (copy_to_user(off_out, &offset, sizeof(loff_t))) 1370a4514ebdSJens Axboe ret = -EFAULT; 1371a4514ebdSJens Axboe 1372a4514ebdSJens Axboe return ret; 1373529565dcSIngo Molnar } 13745274f052SJens Axboe 13757c77f0b3SMiklos Szeredi if (opipe) { 1376529565dcSIngo Molnar if (off_out) 1377529565dcSIngo Molnar return -ESPIPE; 1378b92ce558SJens Axboe if (off_in) { 137919c9a49bSChangli Gao if (!(in->f_mode & FMODE_PREAD)) 1380b92ce558SJens Axboe return -EINVAL; 1381cbb7e577SJens Axboe if (copy_from_user(&offset, off_in, sizeof(loff_t))) 1382b92ce558SJens Axboe return -EFAULT; 13837995bd28SAl Viro } else { 13847995bd28SAl Viro offset = in->f_pos; 13857995bd28SAl Viro } 1386529565dcSIngo Molnar 13877995bd28SAl Viro ret = do_splice_to(in, &offset, opipe, len, flags); 1388a4514ebdSJens Axboe 13897995bd28SAl Viro if (!off_in) 13907995bd28SAl Viro in->f_pos = offset; 13917995bd28SAl Viro else if (copy_to_user(off_in, &offset, sizeof(loff_t))) 1392a4514ebdSJens Axboe ret = -EFAULT; 1393a4514ebdSJens Axboe 1394a4514ebdSJens Axboe return ret; 1395529565dcSIngo Molnar } 13965274f052SJens Axboe 13975274f052SJens Axboe return -EINVAL; 13985274f052SJens Axboe } 13995274f052SJens Axboe 1400912d35f8SJens Axboe /* 1401912d35f8SJens Axboe * Map an iov into an array of pages and offset/length tupples. With the 1402912d35f8SJens Axboe * partial_page structure, we can map several non-contiguous ranges into 1403912d35f8SJens Axboe * our ones pages[] map instead of splitting that operation into pieces. 1404912d35f8SJens Axboe * Could easily be exported as a generic helper for other users, in which 1405912d35f8SJens Axboe * case one would probably want to add a 'max_nr_pages' parameter as well. 1406912d35f8SJens Axboe */ 1407912d35f8SJens Axboe static int get_iovec_page_array(const struct iovec __user *iov, 1408912d35f8SJens Axboe unsigned int nr_vecs, struct page **pages, 1409bd1a68b5SEric Dumazet struct partial_page *partial, bool aligned, 141035f3d14dSJens Axboe unsigned int pipe_buffers) 1411912d35f8SJens Axboe { 1412912d35f8SJens Axboe int buffers = 0, error = 0; 1413912d35f8SJens Axboe 1414912d35f8SJens Axboe while (nr_vecs) { 1415912d35f8SJens Axboe unsigned long off, npages; 141675723957SLinus Torvalds struct iovec entry; 1417912d35f8SJens Axboe void __user *base; 1418912d35f8SJens Axboe size_t len; 1419912d35f8SJens Axboe int i; 1420912d35f8SJens Axboe 142175723957SLinus Torvalds error = -EFAULT; 1422bc40d73cSNick Piggin if (copy_from_user(&entry, iov, sizeof(entry))) 1423912d35f8SJens Axboe break; 142475723957SLinus Torvalds 142575723957SLinus Torvalds base = entry.iov_base; 142675723957SLinus Torvalds len = entry.iov_len; 1427912d35f8SJens Axboe 1428912d35f8SJens Axboe /* 1429912d35f8SJens Axboe * Sanity check this iovec. 0 read succeeds. 1430912d35f8SJens Axboe */ 143175723957SLinus Torvalds error = 0; 1432912d35f8SJens Axboe if (unlikely(!len)) 1433912d35f8SJens Axboe break; 1434912d35f8SJens Axboe error = -EFAULT; 1435712a30e6SBastian Blank if (!access_ok(VERIFY_READ, base, len)) 1436912d35f8SJens Axboe break; 1437912d35f8SJens Axboe 1438912d35f8SJens Axboe /* 1439912d35f8SJens Axboe * Get this base offset and number of pages, then map 1440912d35f8SJens Axboe * in the user pages. 1441912d35f8SJens Axboe */ 1442912d35f8SJens Axboe off = (unsigned long) base & ~PAGE_MASK; 14437afa6fd0SJens Axboe 14447afa6fd0SJens Axboe /* 14457afa6fd0SJens Axboe * If asked for alignment, the offset must be zero and the 14467afa6fd0SJens Axboe * length a multiple of the PAGE_SIZE. 14477afa6fd0SJens Axboe */ 14487afa6fd0SJens Axboe error = -EINVAL; 14497afa6fd0SJens Axboe if (aligned && (off || len & ~PAGE_MASK)) 14507afa6fd0SJens Axboe break; 14517afa6fd0SJens Axboe 1452912d35f8SJens Axboe npages = (off + len + PAGE_SIZE - 1) >> PAGE_SHIFT; 145335f3d14dSJens Axboe if (npages > pipe_buffers - buffers) 145435f3d14dSJens Axboe npages = pipe_buffers - buffers; 1455912d35f8SJens Axboe 1456bc40d73cSNick Piggin error = get_user_pages_fast((unsigned long)base, npages, 1457bc40d73cSNick Piggin 0, &pages[buffers]); 1458912d35f8SJens Axboe 1459912d35f8SJens Axboe if (unlikely(error <= 0)) 1460912d35f8SJens Axboe break; 1461912d35f8SJens Axboe 1462912d35f8SJens Axboe /* 1463912d35f8SJens Axboe * Fill this contiguous range into the partial page map. 1464912d35f8SJens Axboe */ 1465912d35f8SJens Axboe for (i = 0; i < error; i++) { 14667591489aSJens Axboe const int plen = min_t(size_t, len, PAGE_SIZE - off); 1467912d35f8SJens Axboe 1468912d35f8SJens Axboe partial[buffers].offset = off; 1469912d35f8SJens Axboe partial[buffers].len = plen; 1470912d35f8SJens Axboe 1471912d35f8SJens Axboe off = 0; 1472912d35f8SJens Axboe len -= plen; 1473912d35f8SJens Axboe buffers++; 1474912d35f8SJens Axboe } 1475912d35f8SJens Axboe 1476912d35f8SJens Axboe /* 1477912d35f8SJens Axboe * We didn't complete this iov, stop here since it probably 1478912d35f8SJens Axboe * means we have to move some of this into a pipe to 1479912d35f8SJens Axboe * be able to continue. 1480912d35f8SJens Axboe */ 1481912d35f8SJens Axboe if (len) 1482912d35f8SJens Axboe break; 1483912d35f8SJens Axboe 1484912d35f8SJens Axboe /* 1485912d35f8SJens Axboe * Don't continue if we mapped fewer pages than we asked for, 1486912d35f8SJens Axboe * or if we mapped the max number of pages that we have 1487912d35f8SJens Axboe * room for. 1488912d35f8SJens Axboe */ 148935f3d14dSJens Axboe if (error < npages || buffers == pipe_buffers) 1490912d35f8SJens Axboe break; 1491912d35f8SJens Axboe 1492912d35f8SJens Axboe nr_vecs--; 1493912d35f8SJens Axboe iov++; 1494912d35f8SJens Axboe } 1495912d35f8SJens Axboe 1496912d35f8SJens Axboe if (buffers) 1497912d35f8SJens Axboe return buffers; 1498912d35f8SJens Axboe 1499912d35f8SJens Axboe return error; 1500912d35f8SJens Axboe } 1501912d35f8SJens Axboe 15026a14b90bSJens Axboe static int pipe_to_user(struct pipe_inode_info *pipe, struct pipe_buffer *buf, 15036a14b90bSJens Axboe struct splice_desc *sd) 15046a14b90bSJens Axboe { 15056a14b90bSJens Axboe char *src; 15066a14b90bSJens Axboe int ret; 15076a14b90bSJens Axboe 15086a14b90bSJens Axboe /* 15096a14b90bSJens Axboe * See if we can use the atomic maps, by prefaulting in the 15106a14b90bSJens Axboe * pages and doing an atomic copy 15116a14b90bSJens Axboe */ 15126a14b90bSJens Axboe if (!fault_in_pages_writeable(sd->u.userptr, sd->len)) { 15136a14b90bSJens Axboe src = buf->ops->map(pipe, buf, 1); 15146a14b90bSJens Axboe ret = __copy_to_user_inatomic(sd->u.userptr, src + buf->offset, 15156a14b90bSJens Axboe sd->len); 15166a14b90bSJens Axboe buf->ops->unmap(pipe, buf, src); 15176a14b90bSJens Axboe if (!ret) { 15186a14b90bSJens Axboe ret = sd->len; 15196a14b90bSJens Axboe goto out; 15206a14b90bSJens Axboe } 15216a14b90bSJens Axboe } 15226a14b90bSJens Axboe 15236a14b90bSJens Axboe /* 15246a14b90bSJens Axboe * No dice, use slow non-atomic map and copy 15256a14b90bSJens Axboe */ 15266a14b90bSJens Axboe src = buf->ops->map(pipe, buf, 0); 15276a14b90bSJens Axboe 15286a14b90bSJens Axboe ret = sd->len; 15296a14b90bSJens Axboe if (copy_to_user(sd->u.userptr, src + buf->offset, sd->len)) 15306a14b90bSJens Axboe ret = -EFAULT; 15316a14b90bSJens Axboe 15326866bef4SJens Axboe buf->ops->unmap(pipe, buf, src); 15336a14b90bSJens Axboe out: 15346a14b90bSJens Axboe if (ret > 0) 15356a14b90bSJens Axboe sd->u.userptr += ret; 15366a14b90bSJens Axboe return ret; 15376a14b90bSJens Axboe } 15386a14b90bSJens Axboe 15396a14b90bSJens Axboe /* 15406a14b90bSJens Axboe * For lack of a better implementation, implement vmsplice() to userspace 15416a14b90bSJens Axboe * as a simple copy of the pipes pages to the user iov. 15426a14b90bSJens Axboe */ 15436a14b90bSJens Axboe static long vmsplice_to_user(struct file *file, const struct iovec __user *iov, 15446a14b90bSJens Axboe unsigned long nr_segs, unsigned int flags) 15456a14b90bSJens Axboe { 15466a14b90bSJens Axboe struct pipe_inode_info *pipe; 15476a14b90bSJens Axboe struct splice_desc sd; 15486a14b90bSJens Axboe ssize_t size; 15496a14b90bSJens Axboe int error; 15506a14b90bSJens Axboe long ret; 15516a14b90bSJens Axboe 155271993e62SLinus Torvalds pipe = get_pipe_info(file); 15536a14b90bSJens Axboe if (!pipe) 15546a14b90bSJens Axboe return -EBADF; 15556a14b90bSJens Axboe 155661e0d47cSMiklos Szeredi pipe_lock(pipe); 15576a14b90bSJens Axboe 15586a14b90bSJens Axboe error = ret = 0; 15596a14b90bSJens Axboe while (nr_segs) { 15606a14b90bSJens Axboe void __user *base; 15616a14b90bSJens Axboe size_t len; 15626a14b90bSJens Axboe 15636a14b90bSJens Axboe /* 15646a14b90bSJens Axboe * Get user address base and length for this iovec. 15656a14b90bSJens Axboe */ 15666a14b90bSJens Axboe error = get_user(base, &iov->iov_base); 15676a14b90bSJens Axboe if (unlikely(error)) 15686a14b90bSJens Axboe break; 15696a14b90bSJens Axboe error = get_user(len, &iov->iov_len); 15706a14b90bSJens Axboe if (unlikely(error)) 15716a14b90bSJens Axboe break; 15726a14b90bSJens Axboe 15736a14b90bSJens Axboe /* 15746a14b90bSJens Axboe * Sanity check this iovec. 0 read succeeds. 15756a14b90bSJens Axboe */ 15766a14b90bSJens Axboe if (unlikely(!len)) 15776a14b90bSJens Axboe break; 15786a14b90bSJens Axboe if (unlikely(!base)) { 15796a14b90bSJens Axboe error = -EFAULT; 15806a14b90bSJens Axboe break; 15816a14b90bSJens Axboe } 15826a14b90bSJens Axboe 15838811930dSJens Axboe if (unlikely(!access_ok(VERIFY_WRITE, base, len))) { 15848811930dSJens Axboe error = -EFAULT; 15858811930dSJens Axboe break; 15868811930dSJens Axboe } 15878811930dSJens Axboe 15886a14b90bSJens Axboe sd.len = 0; 15896a14b90bSJens Axboe sd.total_len = len; 15906a14b90bSJens Axboe sd.flags = flags; 15916a14b90bSJens Axboe sd.u.userptr = base; 15926a14b90bSJens Axboe sd.pos = 0; 15936a14b90bSJens Axboe 15946a14b90bSJens Axboe size = __splice_from_pipe(pipe, &sd, pipe_to_user); 15956a14b90bSJens Axboe if (size < 0) { 15966a14b90bSJens Axboe if (!ret) 15976a14b90bSJens Axboe ret = size; 15986a14b90bSJens Axboe 15996a14b90bSJens Axboe break; 16006a14b90bSJens Axboe } 16016a14b90bSJens Axboe 16026a14b90bSJens Axboe ret += size; 16036a14b90bSJens Axboe 16046a14b90bSJens Axboe if (size < len) 16056a14b90bSJens Axboe break; 16066a14b90bSJens Axboe 16076a14b90bSJens Axboe nr_segs--; 16086a14b90bSJens Axboe iov++; 16096a14b90bSJens Axboe } 16106a14b90bSJens Axboe 161161e0d47cSMiklos Szeredi pipe_unlock(pipe); 16126a14b90bSJens Axboe 16136a14b90bSJens Axboe if (!ret) 16146a14b90bSJens Axboe ret = error; 16156a14b90bSJens Axboe 16166a14b90bSJens Axboe return ret; 16176a14b90bSJens Axboe } 16186a14b90bSJens Axboe 1619912d35f8SJens Axboe /* 1620912d35f8SJens Axboe * vmsplice splices a user address range into a pipe. It can be thought of 1621912d35f8SJens Axboe * as splice-from-memory, where the regular splice is splice-from-file (or 1622912d35f8SJens Axboe * to file). In both cases the output is a pipe, naturally. 1623912d35f8SJens Axboe */ 16246a14b90bSJens Axboe static long vmsplice_to_pipe(struct file *file, const struct iovec __user *iov, 1625912d35f8SJens Axboe unsigned long nr_segs, unsigned int flags) 1626912d35f8SJens Axboe { 1627ddac0d39SJens Axboe struct pipe_inode_info *pipe; 162835f3d14dSJens Axboe struct page *pages[PIPE_DEF_BUFFERS]; 162935f3d14dSJens Axboe struct partial_page partial[PIPE_DEF_BUFFERS]; 1630912d35f8SJens Axboe struct splice_pipe_desc spd = { 1631912d35f8SJens Axboe .pages = pages, 1632912d35f8SJens Axboe .partial = partial, 1633047fe360SEric Dumazet .nr_pages_max = PIPE_DEF_BUFFERS, 1634912d35f8SJens Axboe .flags = flags, 1635912d35f8SJens Axboe .ops = &user_page_pipe_buf_ops, 1636bbdfc2f7SJens Axboe .spd_release = spd_release_page, 1637912d35f8SJens Axboe }; 163835f3d14dSJens Axboe long ret; 1639912d35f8SJens Axboe 164071993e62SLinus Torvalds pipe = get_pipe_info(file); 1641ddac0d39SJens Axboe if (!pipe) 1642912d35f8SJens Axboe return -EBADF; 1643912d35f8SJens Axboe 164435f3d14dSJens Axboe if (splice_grow_spd(pipe, &spd)) 164535f3d14dSJens Axboe return -ENOMEM; 1646912d35f8SJens Axboe 164735f3d14dSJens Axboe spd.nr_pages = get_iovec_page_array(iov, nr_segs, spd.pages, 1648bd1a68b5SEric Dumazet spd.partial, false, 1649047fe360SEric Dumazet spd.nr_pages_max); 165035f3d14dSJens Axboe if (spd.nr_pages <= 0) 165135f3d14dSJens Axboe ret = spd.nr_pages; 165235f3d14dSJens Axboe else 165335f3d14dSJens Axboe ret = splice_to_pipe(pipe, &spd); 165435f3d14dSJens Axboe 1655047fe360SEric Dumazet splice_shrink_spd(&spd); 165635f3d14dSJens Axboe return ret; 1657912d35f8SJens Axboe } 1658912d35f8SJens Axboe 16596a14b90bSJens Axboe /* 16606a14b90bSJens Axboe * Note that vmsplice only really supports true splicing _from_ user memory 16616a14b90bSJens Axboe * to a pipe, not the other way around. Splicing from user memory is a simple 16626a14b90bSJens Axboe * operation that can be supported without any funky alignment restrictions 16636a14b90bSJens Axboe * or nasty vm tricks. We simply map in the user memory and fill them into 16646a14b90bSJens Axboe * a pipe. The reverse isn't quite as easy, though. There are two possible 16656a14b90bSJens Axboe * solutions for that: 16666a14b90bSJens Axboe * 16676a14b90bSJens Axboe * - memcpy() the data internally, at which point we might as well just 16686a14b90bSJens Axboe * do a regular read() on the buffer anyway. 16696a14b90bSJens Axboe * - Lots of nasty vm tricks, that are neither fast nor flexible (it 16706a14b90bSJens Axboe * has restriction limitations on both ends of the pipe). 16716a14b90bSJens Axboe * 16726a14b90bSJens Axboe * Currently we punt and implement it as a normal copy, see pipe_to_user(). 16736a14b90bSJens Axboe * 16746a14b90bSJens Axboe */ 1675836f92adSHeiko Carstens SYSCALL_DEFINE4(vmsplice, int, fd, const struct iovec __user *, iov, 1676836f92adSHeiko Carstens unsigned long, nr_segs, unsigned int, flags) 1677912d35f8SJens Axboe { 16782903ff01SAl Viro struct fd f; 1679912d35f8SJens Axboe long error; 1680912d35f8SJens Axboe 16816a14b90bSJens Axboe if (unlikely(nr_segs > UIO_MAXIOV)) 16826a14b90bSJens Axboe return -EINVAL; 16836a14b90bSJens Axboe else if (unlikely(!nr_segs)) 16846a14b90bSJens Axboe return 0; 16856a14b90bSJens Axboe 1686912d35f8SJens Axboe error = -EBADF; 16872903ff01SAl Viro f = fdget(fd); 16882903ff01SAl Viro if (f.file) { 16892903ff01SAl Viro if (f.file->f_mode & FMODE_WRITE) 16902903ff01SAl Viro error = vmsplice_to_pipe(f.file, iov, nr_segs, flags); 16912903ff01SAl Viro else if (f.file->f_mode & FMODE_READ) 16922903ff01SAl Viro error = vmsplice_to_user(f.file, iov, nr_segs, flags); 1693912d35f8SJens Axboe 16942903ff01SAl Viro fdput(f); 1695912d35f8SJens Axboe } 1696912d35f8SJens Axboe 1697912d35f8SJens Axboe return error; 1698912d35f8SJens Axboe } 1699912d35f8SJens Axboe 170076b021d0SAl Viro #ifdef CONFIG_COMPAT 170176b021d0SAl Viro COMPAT_SYSCALL_DEFINE4(vmsplice, int, fd, const struct compat_iovec __user *, iov32, 170276b021d0SAl Viro unsigned int, nr_segs, unsigned int, flags) 170376b021d0SAl Viro { 170476b021d0SAl Viro unsigned i; 170576b021d0SAl Viro struct iovec __user *iov; 170676b021d0SAl Viro if (nr_segs > UIO_MAXIOV) 170776b021d0SAl Viro return -EINVAL; 170876b021d0SAl Viro iov = compat_alloc_user_space(nr_segs * sizeof(struct iovec)); 170976b021d0SAl Viro for (i = 0; i < nr_segs; i++) { 171076b021d0SAl Viro struct compat_iovec v; 171176b021d0SAl Viro if (get_user(v.iov_base, &iov32[i].iov_base) || 171276b021d0SAl Viro get_user(v.iov_len, &iov32[i].iov_len) || 171376b021d0SAl Viro put_user(compat_ptr(v.iov_base), &iov[i].iov_base) || 171476b021d0SAl Viro put_user(v.iov_len, &iov[i].iov_len)) 171576b021d0SAl Viro return -EFAULT; 171676b021d0SAl Viro } 171776b021d0SAl Viro return sys_vmsplice(fd, iov, nr_segs, flags); 171876b021d0SAl Viro } 171976b021d0SAl Viro #endif 172076b021d0SAl Viro 1721836f92adSHeiko Carstens SYSCALL_DEFINE6(splice, int, fd_in, loff_t __user *, off_in, 1722836f92adSHeiko Carstens int, fd_out, loff_t __user *, off_out, 1723836f92adSHeiko Carstens size_t, len, unsigned int, flags) 17245274f052SJens Axboe { 17252903ff01SAl Viro struct fd in, out; 17265274f052SJens Axboe long error; 17275274f052SJens Axboe 17285274f052SJens Axboe if (unlikely(!len)) 17295274f052SJens Axboe return 0; 17305274f052SJens Axboe 17315274f052SJens Axboe error = -EBADF; 17322903ff01SAl Viro in = fdget(fd_in); 17332903ff01SAl Viro if (in.file) { 17342903ff01SAl Viro if (in.file->f_mode & FMODE_READ) { 17352903ff01SAl Viro out = fdget(fd_out); 17362903ff01SAl Viro if (out.file) { 17372903ff01SAl Viro if (out.file->f_mode & FMODE_WRITE) 17382903ff01SAl Viro error = do_splice(in.file, off_in, 17392903ff01SAl Viro out.file, off_out, 1740529565dcSIngo Molnar len, flags); 17412903ff01SAl Viro fdput(out); 17425274f052SJens Axboe } 17435274f052SJens Axboe } 17442903ff01SAl Viro fdput(in); 17455274f052SJens Axboe } 17465274f052SJens Axboe return error; 17475274f052SJens Axboe } 174870524490SJens Axboe 174970524490SJens Axboe /* 1750aadd06e5SJens Axboe * Make sure there's data to read. Wait for input if we can, otherwise 1751aadd06e5SJens Axboe * return an appropriate error. 1752aadd06e5SJens Axboe */ 17537c77f0b3SMiklos Szeredi static int ipipe_prep(struct pipe_inode_info *pipe, unsigned int flags) 1754aadd06e5SJens Axboe { 1755aadd06e5SJens Axboe int ret; 1756aadd06e5SJens Axboe 1757aadd06e5SJens Axboe /* 1758aadd06e5SJens Axboe * Check ->nrbufs without the inode lock first. This function 1759aadd06e5SJens Axboe * is speculative anyways, so missing one is ok. 1760aadd06e5SJens Axboe */ 1761aadd06e5SJens Axboe if (pipe->nrbufs) 1762aadd06e5SJens Axboe return 0; 1763aadd06e5SJens Axboe 1764aadd06e5SJens Axboe ret = 0; 176561e0d47cSMiklos Szeredi pipe_lock(pipe); 1766aadd06e5SJens Axboe 1767aadd06e5SJens Axboe while (!pipe->nrbufs) { 1768aadd06e5SJens Axboe if (signal_pending(current)) { 1769aadd06e5SJens Axboe ret = -ERESTARTSYS; 1770aadd06e5SJens Axboe break; 1771aadd06e5SJens Axboe } 1772aadd06e5SJens Axboe if (!pipe->writers) 1773aadd06e5SJens Axboe break; 1774aadd06e5SJens Axboe if (!pipe->waiting_writers) { 1775aadd06e5SJens Axboe if (flags & SPLICE_F_NONBLOCK) { 1776aadd06e5SJens Axboe ret = -EAGAIN; 1777aadd06e5SJens Axboe break; 1778aadd06e5SJens Axboe } 1779aadd06e5SJens Axboe } 1780aadd06e5SJens Axboe pipe_wait(pipe); 1781aadd06e5SJens Axboe } 1782aadd06e5SJens Axboe 178361e0d47cSMiklos Szeredi pipe_unlock(pipe); 1784aadd06e5SJens Axboe return ret; 1785aadd06e5SJens Axboe } 1786aadd06e5SJens Axboe 1787aadd06e5SJens Axboe /* 1788aadd06e5SJens Axboe * Make sure there's writeable room. Wait for room if we can, otherwise 1789aadd06e5SJens Axboe * return an appropriate error. 1790aadd06e5SJens Axboe */ 17917c77f0b3SMiklos Szeredi static int opipe_prep(struct pipe_inode_info *pipe, unsigned int flags) 1792aadd06e5SJens Axboe { 1793aadd06e5SJens Axboe int ret; 1794aadd06e5SJens Axboe 1795aadd06e5SJens Axboe /* 1796aadd06e5SJens Axboe * Check ->nrbufs without the inode lock first. This function 1797aadd06e5SJens Axboe * is speculative anyways, so missing one is ok. 1798aadd06e5SJens Axboe */ 179935f3d14dSJens Axboe if (pipe->nrbufs < pipe->buffers) 1800aadd06e5SJens Axboe return 0; 1801aadd06e5SJens Axboe 1802aadd06e5SJens Axboe ret = 0; 180361e0d47cSMiklos Szeredi pipe_lock(pipe); 1804aadd06e5SJens Axboe 180535f3d14dSJens Axboe while (pipe->nrbufs >= pipe->buffers) { 1806aadd06e5SJens Axboe if (!pipe->readers) { 1807aadd06e5SJens Axboe send_sig(SIGPIPE, current, 0); 1808aadd06e5SJens Axboe ret = -EPIPE; 1809aadd06e5SJens Axboe break; 1810aadd06e5SJens Axboe } 1811aadd06e5SJens Axboe if (flags & SPLICE_F_NONBLOCK) { 1812aadd06e5SJens Axboe ret = -EAGAIN; 1813aadd06e5SJens Axboe break; 1814aadd06e5SJens Axboe } 1815aadd06e5SJens Axboe if (signal_pending(current)) { 1816aadd06e5SJens Axboe ret = -ERESTARTSYS; 1817aadd06e5SJens Axboe break; 1818aadd06e5SJens Axboe } 1819aadd06e5SJens Axboe pipe->waiting_writers++; 1820aadd06e5SJens Axboe pipe_wait(pipe); 1821aadd06e5SJens Axboe pipe->waiting_writers--; 1822aadd06e5SJens Axboe } 1823aadd06e5SJens Axboe 182461e0d47cSMiklos Szeredi pipe_unlock(pipe); 1825aadd06e5SJens Axboe return ret; 1826aadd06e5SJens Axboe } 1827aadd06e5SJens Axboe 1828aadd06e5SJens Axboe /* 18297c77f0b3SMiklos Szeredi * Splice contents of ipipe to opipe. 18307c77f0b3SMiklos Szeredi */ 18317c77f0b3SMiklos Szeredi static int splice_pipe_to_pipe(struct pipe_inode_info *ipipe, 18327c77f0b3SMiklos Szeredi struct pipe_inode_info *opipe, 18337c77f0b3SMiklos Szeredi size_t len, unsigned int flags) 18347c77f0b3SMiklos Szeredi { 18357c77f0b3SMiklos Szeredi struct pipe_buffer *ibuf, *obuf; 18367c77f0b3SMiklos Szeredi int ret = 0, nbuf; 18377c77f0b3SMiklos Szeredi bool input_wakeup = false; 18387c77f0b3SMiklos Szeredi 18397c77f0b3SMiklos Szeredi 18407c77f0b3SMiklos Szeredi retry: 18417c77f0b3SMiklos Szeredi ret = ipipe_prep(ipipe, flags); 18427c77f0b3SMiklos Szeredi if (ret) 18437c77f0b3SMiklos Szeredi return ret; 18447c77f0b3SMiklos Szeredi 18457c77f0b3SMiklos Szeredi ret = opipe_prep(opipe, flags); 18467c77f0b3SMiklos Szeredi if (ret) 18477c77f0b3SMiklos Szeredi return ret; 18487c77f0b3SMiklos Szeredi 18497c77f0b3SMiklos Szeredi /* 18507c77f0b3SMiklos Szeredi * Potential ABBA deadlock, work around it by ordering lock 18517c77f0b3SMiklos Szeredi * grabbing by pipe info address. Otherwise two different processes 18527c77f0b3SMiklos Szeredi * could deadlock (one doing tee from A -> B, the other from B -> A). 18537c77f0b3SMiklos Szeredi */ 18547c77f0b3SMiklos Szeredi pipe_double_lock(ipipe, opipe); 18557c77f0b3SMiklos Szeredi 18567c77f0b3SMiklos Szeredi do { 18577c77f0b3SMiklos Szeredi if (!opipe->readers) { 18587c77f0b3SMiklos Szeredi send_sig(SIGPIPE, current, 0); 18597c77f0b3SMiklos Szeredi if (!ret) 18607c77f0b3SMiklos Szeredi ret = -EPIPE; 18617c77f0b3SMiklos Szeredi break; 18627c77f0b3SMiklos Szeredi } 18637c77f0b3SMiklos Szeredi 18647c77f0b3SMiklos Szeredi if (!ipipe->nrbufs && !ipipe->writers) 18657c77f0b3SMiklos Szeredi break; 18667c77f0b3SMiklos Szeredi 18677c77f0b3SMiklos Szeredi /* 18687c77f0b3SMiklos Szeredi * Cannot make any progress, because either the input 18697c77f0b3SMiklos Szeredi * pipe is empty or the output pipe is full. 18707c77f0b3SMiklos Szeredi */ 187135f3d14dSJens Axboe if (!ipipe->nrbufs || opipe->nrbufs >= opipe->buffers) { 18727c77f0b3SMiklos Szeredi /* Already processed some buffers, break */ 18737c77f0b3SMiklos Szeredi if (ret) 18747c77f0b3SMiklos Szeredi break; 18757c77f0b3SMiklos Szeredi 18767c77f0b3SMiklos Szeredi if (flags & SPLICE_F_NONBLOCK) { 18777c77f0b3SMiklos Szeredi ret = -EAGAIN; 18787c77f0b3SMiklos Szeredi break; 18797c77f0b3SMiklos Szeredi } 18807c77f0b3SMiklos Szeredi 18817c77f0b3SMiklos Szeredi /* 18827c77f0b3SMiklos Szeredi * We raced with another reader/writer and haven't 18837c77f0b3SMiklos Szeredi * managed to process any buffers. A zero return 18847c77f0b3SMiklos Szeredi * value means EOF, so retry instead. 18857c77f0b3SMiklos Szeredi */ 18867c77f0b3SMiklos Szeredi pipe_unlock(ipipe); 18877c77f0b3SMiklos Szeredi pipe_unlock(opipe); 18887c77f0b3SMiklos Szeredi goto retry; 18897c77f0b3SMiklos Szeredi } 18907c77f0b3SMiklos Szeredi 18917c77f0b3SMiklos Szeredi ibuf = ipipe->bufs + ipipe->curbuf; 189235f3d14dSJens Axboe nbuf = (opipe->curbuf + opipe->nrbufs) & (opipe->buffers - 1); 18937c77f0b3SMiklos Szeredi obuf = opipe->bufs + nbuf; 18947c77f0b3SMiklos Szeredi 18957c77f0b3SMiklos Szeredi if (len >= ibuf->len) { 18967c77f0b3SMiklos Szeredi /* 18977c77f0b3SMiklos Szeredi * Simply move the whole buffer from ipipe to opipe 18987c77f0b3SMiklos Szeredi */ 18997c77f0b3SMiklos Szeredi *obuf = *ibuf; 19007c77f0b3SMiklos Szeredi ibuf->ops = NULL; 19017c77f0b3SMiklos Szeredi opipe->nrbufs++; 190235f3d14dSJens Axboe ipipe->curbuf = (ipipe->curbuf + 1) & (ipipe->buffers - 1); 19037c77f0b3SMiklos Szeredi ipipe->nrbufs--; 19047c77f0b3SMiklos Szeredi input_wakeup = true; 19057c77f0b3SMiklos Szeredi } else { 19067c77f0b3SMiklos Szeredi /* 19077c77f0b3SMiklos Szeredi * Get a reference to this pipe buffer, 19087c77f0b3SMiklos Szeredi * so we can copy the contents over. 19097c77f0b3SMiklos Szeredi */ 19107c77f0b3SMiklos Szeredi ibuf->ops->get(ipipe, ibuf); 19117c77f0b3SMiklos Szeredi *obuf = *ibuf; 19127c77f0b3SMiklos Szeredi 19137c77f0b3SMiklos Szeredi /* 19147c77f0b3SMiklos Szeredi * Don't inherit the gift flag, we need to 19157c77f0b3SMiklos Szeredi * prevent multiple steals of this page. 19167c77f0b3SMiklos Szeredi */ 19177c77f0b3SMiklos Szeredi obuf->flags &= ~PIPE_BUF_FLAG_GIFT; 19187c77f0b3SMiklos Szeredi 19197c77f0b3SMiklos Szeredi obuf->len = len; 19207c77f0b3SMiklos Szeredi opipe->nrbufs++; 19217c77f0b3SMiklos Szeredi ibuf->offset += obuf->len; 19227c77f0b3SMiklos Szeredi ibuf->len -= obuf->len; 19237c77f0b3SMiklos Szeredi } 19247c77f0b3SMiklos Szeredi ret += obuf->len; 19257c77f0b3SMiklos Szeredi len -= obuf->len; 19267c77f0b3SMiklos Szeredi } while (len); 19277c77f0b3SMiklos Szeredi 19287c77f0b3SMiklos Szeredi pipe_unlock(ipipe); 19297c77f0b3SMiklos Szeredi pipe_unlock(opipe); 19307c77f0b3SMiklos Szeredi 19317c77f0b3SMiklos Szeredi /* 19327c77f0b3SMiklos Szeredi * If we put data in the output pipe, wakeup any potential readers. 19337c77f0b3SMiklos Szeredi */ 1934825cdcb1SNamhyung Kim if (ret > 0) 1935825cdcb1SNamhyung Kim wakeup_pipe_readers(opipe); 1936825cdcb1SNamhyung Kim 19377c77f0b3SMiklos Szeredi if (input_wakeup) 19387c77f0b3SMiklos Szeredi wakeup_pipe_writers(ipipe); 19397c77f0b3SMiklos Szeredi 19407c77f0b3SMiklos Szeredi return ret; 19417c77f0b3SMiklos Szeredi } 19427c77f0b3SMiklos Szeredi 19437c77f0b3SMiklos Szeredi /* 194470524490SJens Axboe * Link contents of ipipe to opipe. 194570524490SJens Axboe */ 194670524490SJens Axboe static int link_pipe(struct pipe_inode_info *ipipe, 194770524490SJens Axboe struct pipe_inode_info *opipe, 194870524490SJens Axboe size_t len, unsigned int flags) 194970524490SJens Axboe { 195070524490SJens Axboe struct pipe_buffer *ibuf, *obuf; 1951aadd06e5SJens Axboe int ret = 0, i = 0, nbuf; 195270524490SJens Axboe 195370524490SJens Axboe /* 195470524490SJens Axboe * Potential ABBA deadlock, work around it by ordering lock 195561e0d47cSMiklos Szeredi * grabbing by pipe info address. Otherwise two different processes 195670524490SJens Axboe * could deadlock (one doing tee from A -> B, the other from B -> A). 195770524490SJens Axboe */ 195861e0d47cSMiklos Szeredi pipe_double_lock(ipipe, opipe); 195970524490SJens Axboe 1960aadd06e5SJens Axboe do { 196170524490SJens Axboe if (!opipe->readers) { 196270524490SJens Axboe send_sig(SIGPIPE, current, 0); 196370524490SJens Axboe if (!ret) 196470524490SJens Axboe ret = -EPIPE; 196570524490SJens Axboe break; 196670524490SJens Axboe } 196770524490SJens Axboe 196870524490SJens Axboe /* 1969aadd06e5SJens Axboe * If we have iterated all input buffers or ran out of 1970aadd06e5SJens Axboe * output room, break. 197170524490SJens Axboe */ 197235f3d14dSJens Axboe if (i >= ipipe->nrbufs || opipe->nrbufs >= opipe->buffers) 1973aadd06e5SJens Axboe break; 1974aadd06e5SJens Axboe 197535f3d14dSJens Axboe ibuf = ipipe->bufs + ((ipipe->curbuf + i) & (ipipe->buffers-1)); 197635f3d14dSJens Axboe nbuf = (opipe->curbuf + opipe->nrbufs) & (opipe->buffers - 1); 197770524490SJens Axboe 197870524490SJens Axboe /* 197970524490SJens Axboe * Get a reference to this pipe buffer, 198070524490SJens Axboe * so we can copy the contents over. 198170524490SJens Axboe */ 198270524490SJens Axboe ibuf->ops->get(ipipe, ibuf); 198370524490SJens Axboe 198470524490SJens Axboe obuf = opipe->bufs + nbuf; 198570524490SJens Axboe *obuf = *ibuf; 198670524490SJens Axboe 19877afa6fd0SJens Axboe /* 19887afa6fd0SJens Axboe * Don't inherit the gift flag, we need to 19897afa6fd0SJens Axboe * prevent multiple steals of this page. 19907afa6fd0SJens Axboe */ 19917afa6fd0SJens Axboe obuf->flags &= ~PIPE_BUF_FLAG_GIFT; 19927afa6fd0SJens Axboe 199370524490SJens Axboe if (obuf->len > len) 199470524490SJens Axboe obuf->len = len; 199570524490SJens Axboe 199670524490SJens Axboe opipe->nrbufs++; 199770524490SJens Axboe ret += obuf->len; 199870524490SJens Axboe len -= obuf->len; 1999aadd06e5SJens Axboe i++; 2000aadd06e5SJens Axboe } while (len); 200170524490SJens Axboe 200202cf01aeSJens Axboe /* 200302cf01aeSJens Axboe * return EAGAIN if we have the potential of some data in the 200402cf01aeSJens Axboe * future, otherwise just return 0 200502cf01aeSJens Axboe */ 200602cf01aeSJens Axboe if (!ret && ipipe->waiting_writers && (flags & SPLICE_F_NONBLOCK)) 200702cf01aeSJens Axboe ret = -EAGAIN; 200802cf01aeSJens Axboe 200961e0d47cSMiklos Szeredi pipe_unlock(ipipe); 201061e0d47cSMiklos Szeredi pipe_unlock(opipe); 201170524490SJens Axboe 2012aadd06e5SJens Axboe /* 2013aadd06e5SJens Axboe * If we put data in the output pipe, wakeup any potential readers. 2014aadd06e5SJens Axboe */ 2015825cdcb1SNamhyung Kim if (ret > 0) 2016825cdcb1SNamhyung Kim wakeup_pipe_readers(opipe); 201770524490SJens Axboe 201870524490SJens Axboe return ret; 201970524490SJens Axboe } 202070524490SJens Axboe 202170524490SJens Axboe /* 202270524490SJens Axboe * This is a tee(1) implementation that works on pipes. It doesn't copy 202370524490SJens Axboe * any data, it simply references the 'in' pages on the 'out' pipe. 202470524490SJens Axboe * The 'flags' used are the SPLICE_F_* variants, currently the only 202570524490SJens Axboe * applicable one is SPLICE_F_NONBLOCK. 202670524490SJens Axboe */ 202770524490SJens Axboe static long do_tee(struct file *in, struct file *out, size_t len, 202870524490SJens Axboe unsigned int flags) 202970524490SJens Axboe { 203071993e62SLinus Torvalds struct pipe_inode_info *ipipe = get_pipe_info(in); 203171993e62SLinus Torvalds struct pipe_inode_info *opipe = get_pipe_info(out); 2032aadd06e5SJens Axboe int ret = -EINVAL; 203370524490SJens Axboe 203470524490SJens Axboe /* 2035aadd06e5SJens Axboe * Duplicate the contents of ipipe to opipe without actually 2036aadd06e5SJens Axboe * copying the data. 203770524490SJens Axboe */ 2038aadd06e5SJens Axboe if (ipipe && opipe && ipipe != opipe) { 2039aadd06e5SJens Axboe /* 2040aadd06e5SJens Axboe * Keep going, unless we encounter an error. The ipipe/opipe 2041aadd06e5SJens Axboe * ordering doesn't really matter. 2042aadd06e5SJens Axboe */ 20437c77f0b3SMiklos Szeredi ret = ipipe_prep(ipipe, flags); 2044aadd06e5SJens Axboe if (!ret) { 20457c77f0b3SMiklos Szeredi ret = opipe_prep(opipe, flags); 204602cf01aeSJens Axboe if (!ret) 2047aadd06e5SJens Axboe ret = link_pipe(ipipe, opipe, len, flags); 2048aadd06e5SJens Axboe } 2049aadd06e5SJens Axboe } 205070524490SJens Axboe 2051aadd06e5SJens Axboe return ret; 205270524490SJens Axboe } 205370524490SJens Axboe 2054836f92adSHeiko Carstens SYSCALL_DEFINE4(tee, int, fdin, int, fdout, size_t, len, unsigned int, flags) 205570524490SJens Axboe { 20562903ff01SAl Viro struct fd in; 20572903ff01SAl Viro int error; 205870524490SJens Axboe 205970524490SJens Axboe if (unlikely(!len)) 206070524490SJens Axboe return 0; 206170524490SJens Axboe 206270524490SJens Axboe error = -EBADF; 20632903ff01SAl Viro in = fdget(fdin); 20642903ff01SAl Viro if (in.file) { 20652903ff01SAl Viro if (in.file->f_mode & FMODE_READ) { 20662903ff01SAl Viro struct fd out = fdget(fdout); 20672903ff01SAl Viro if (out.file) { 20682903ff01SAl Viro if (out.file->f_mode & FMODE_WRITE) 20692903ff01SAl Viro error = do_tee(in.file, out.file, 20702903ff01SAl Viro len, flags); 20712903ff01SAl Viro fdput(out); 207270524490SJens Axboe } 207370524490SJens Axboe } 20742903ff01SAl Viro fdput(in); 207570524490SJens Axboe } 207670524490SJens Axboe 207770524490SJens Axboe return error; 207870524490SJens Axboe } 2079