1457c8996SThomas Gleixner // SPDX-License-Identifier: GPL-2.0-only 25274f052SJens Axboe /* 35274f052SJens Axboe * "splice": joining two ropes together by interweaving their strands. 45274f052SJens Axboe * 55274f052SJens Axboe * This is the "extended pipe" functionality, where a pipe is used as 65274f052SJens Axboe * an arbitrary in-memory buffer. Think of a pipe as a small kernel 75274f052SJens Axboe * buffer that you can use to transfer data from one end to the other. 85274f052SJens Axboe * 95274f052SJens Axboe * The traditional unix read/write is extended with a "splice()" operation 105274f052SJens Axboe * that transfers data buffers to or from a pipe buffer. 115274f052SJens Axboe * 125274f052SJens Axboe * Named by Larry McVoy, original implementation from Linus, extended by 13c2058e06SJens Axboe * Jens to support splicing to files, network, direct splicing, etc and 14c2058e06SJens Axboe * fixing lots of bugs. 155274f052SJens Axboe * 160fe23479SJens Axboe * Copyright (C) 2005-2006 Jens Axboe <axboe@kernel.dk> 17c2058e06SJens Axboe * Copyright (C) 2005-2006 Linus Torvalds <torvalds@osdl.org> 18c2058e06SJens Axboe * Copyright (C) 2006 Ingo Molnar <mingo@elte.hu> 195274f052SJens Axboe * 205274f052SJens Axboe */ 21be297968SChristoph Hellwig #include <linux/bvec.h> 225274f052SJens Axboe #include <linux/fs.h> 235274f052SJens Axboe #include <linux/file.h> 245274f052SJens Axboe #include <linux/pagemap.h> 25d6b29d7cSJens Axboe #include <linux/splice.h> 2608e552c6SKAMEZAWA Hiroyuki #include <linux/memcontrol.h> 275274f052SJens Axboe #include <linux/mm_inline.h> 285abc97aaSJens Axboe #include <linux/swap.h> 294f6f0bd2SJens Axboe #include <linux/writeback.h> 30630d9c47SPaul Gortmaker #include <linux/export.h> 314f6f0bd2SJens Axboe #include <linux/syscalls.h> 32912d35f8SJens Axboe #include <linux/uio.h> 3329ce2058SJames Morris #include <linux/security.h> 345a0e3ad6STejun Heo #include <linux/gfp.h> 3535f9c09fSEric Dumazet #include <linux/socket.h> 3676b021d0SAl Viro #include <linux/compat.h> 37174cd4b1SIngo Molnar #include <linux/sched/signal.h> 38174cd4b1SIngo Molnar 3906ae43f3SAl Viro #include "internal.h" 405274f052SJens Axboe 4183f9135bSJens Axboe /* 4283f9135bSJens Axboe * Attempt to steal a page from a pipe buffer. This should perhaps go into 4383f9135bSJens Axboe * a vm helper function, it's already simplified quite a bit by the 4483f9135bSJens Axboe * addition of remove_mapping(). If success is returned, the caller may 4583f9135bSJens Axboe * attempt to reuse this page for another destination. 4683f9135bSJens Axboe */ 4776ad4d11SJens Axboe static int page_cache_pipe_buf_steal(struct pipe_inode_info *pipe, 485abc97aaSJens Axboe struct pipe_buffer *buf) 495abc97aaSJens Axboe { 505abc97aaSJens Axboe struct page *page = buf->page; 519e94cd4fSJens Axboe struct address_space *mapping; 525abc97aaSJens Axboe 539e0267c2SJens Axboe lock_page(page); 549e0267c2SJens Axboe 559e94cd4fSJens Axboe mapping = page_mapping(page); 569e94cd4fSJens Axboe if (mapping) { 575abc97aaSJens Axboe WARN_ON(!PageUptodate(page)); 585abc97aaSJens Axboe 59ad8d6f0aSJens Axboe /* 609e94cd4fSJens Axboe * At least for ext2 with nobh option, we need to wait on 619e94cd4fSJens Axboe * writeback completing on this page, since we'll remove it 629e94cd4fSJens Axboe * from the pagecache. Otherwise truncate wont wait on the 639e94cd4fSJens Axboe * page, allowing the disk blocks to be reused by someone else 649e94cd4fSJens Axboe * before we actually wrote our data to them. fs corruption 659e94cd4fSJens Axboe * ensues. 66ad8d6f0aSJens Axboe */ 67ad8d6f0aSJens Axboe wait_on_page_writeback(page); 68ad8d6f0aSJens Axboe 69266cf658SDavid Howells if (page_has_private(page) && 70266cf658SDavid Howells !try_to_release_page(page, GFP_KERNEL)) 71ca39d651SJens Axboe goto out_unlock; 724f6f0bd2SJens Axboe 739e94cd4fSJens Axboe /* 749e94cd4fSJens Axboe * If we succeeded in removing the mapping, set LRU flag 759e94cd4fSJens Axboe * and return good. 769e94cd4fSJens Axboe */ 779e94cd4fSJens Axboe if (remove_mapping(mapping, page)) { 781432873aSJens Axboe buf->flags |= PIPE_BUF_FLAG_LRU; 795abc97aaSJens Axboe return 0; 805abc97aaSJens Axboe } 819e94cd4fSJens Axboe } 829e94cd4fSJens Axboe 839e94cd4fSJens Axboe /* 849e94cd4fSJens Axboe * Raced with truncate or failed to remove page from current 859e94cd4fSJens Axboe * address space, unlock and return failure. 869e94cd4fSJens Axboe */ 87ca39d651SJens Axboe out_unlock: 889e94cd4fSJens Axboe unlock_page(page); 899e94cd4fSJens Axboe return 1; 909e94cd4fSJens Axboe } 915abc97aaSJens Axboe 9276ad4d11SJens Axboe static void page_cache_pipe_buf_release(struct pipe_inode_info *pipe, 935274f052SJens Axboe struct pipe_buffer *buf) 945274f052SJens Axboe { 9509cbfeafSKirill A. Shutemov put_page(buf->page); 961432873aSJens Axboe buf->flags &= ~PIPE_BUF_FLAG_LRU; 975274f052SJens Axboe } 985274f052SJens Axboe 990845718dSJens Axboe /* 1000845718dSJens Axboe * Check whether the contents of buf is OK to access. Since the content 1010845718dSJens Axboe * is a page cache page, IO may be in flight. 1020845718dSJens Axboe */ 103cac36bb0SJens Axboe static int page_cache_pipe_buf_confirm(struct pipe_inode_info *pipe, 1045274f052SJens Axboe struct pipe_buffer *buf) 1055274f052SJens Axboe { 1065274f052SJens Axboe struct page *page = buf->page; 10749d0b21bSJens Axboe int err; 1085274f052SJens Axboe 1095274f052SJens Axboe if (!PageUptodate(page)) { 11049d0b21bSJens Axboe lock_page(page); 1115274f052SJens Axboe 11249d0b21bSJens Axboe /* 11349d0b21bSJens Axboe * Page got truncated/unhashed. This will cause a 0-byte 11473d62d83SIngo Molnar * splice, if this is the first page. 11549d0b21bSJens Axboe */ 1165274f052SJens Axboe if (!page->mapping) { 11749d0b21bSJens Axboe err = -ENODATA; 11849d0b21bSJens Axboe goto error; 1195274f052SJens Axboe } 1205274f052SJens Axboe 12149d0b21bSJens Axboe /* 12273d62d83SIngo Molnar * Uh oh, read-error from disk. 12349d0b21bSJens Axboe */ 12449d0b21bSJens Axboe if (!PageUptodate(page)) { 12549d0b21bSJens Axboe err = -EIO; 12649d0b21bSJens Axboe goto error; 12749d0b21bSJens Axboe } 12849d0b21bSJens Axboe 12949d0b21bSJens Axboe /* 130f84d7519SJens Axboe * Page is ok afterall, we are done. 13149d0b21bSJens Axboe */ 13249d0b21bSJens Axboe unlock_page(page); 13349d0b21bSJens Axboe } 13449d0b21bSJens Axboe 135f84d7519SJens Axboe return 0; 13649d0b21bSJens Axboe error: 13749d0b21bSJens Axboe unlock_page(page); 138f84d7519SJens Axboe return err; 13970524490SJens Axboe } 14070524490SJens Axboe 141708e3508SHugh Dickins const struct pipe_buf_operations page_cache_pipe_buf_ops = { 142cac36bb0SJens Axboe .confirm = page_cache_pipe_buf_confirm, 1435274f052SJens Axboe .release = page_cache_pipe_buf_release, 1445abc97aaSJens Axboe .steal = page_cache_pipe_buf_steal, 145f84d7519SJens Axboe .get = generic_pipe_buf_get, 1465274f052SJens Axboe }; 1475274f052SJens Axboe 148912d35f8SJens Axboe static int user_page_pipe_buf_steal(struct pipe_inode_info *pipe, 149912d35f8SJens Axboe struct pipe_buffer *buf) 150912d35f8SJens Axboe { 1517afa6fd0SJens Axboe if (!(buf->flags & PIPE_BUF_FLAG_GIFT)) 152912d35f8SJens Axboe return 1; 1537afa6fd0SJens Axboe 1541432873aSJens Axboe buf->flags |= PIPE_BUF_FLAG_LRU; 155330ab716SJens Axboe return generic_pipe_buf_steal(pipe, buf); 156912d35f8SJens Axboe } 157912d35f8SJens Axboe 158d4c3cca9SEric Dumazet static const struct pipe_buf_operations user_page_pipe_buf_ops = { 159cac36bb0SJens Axboe .confirm = generic_pipe_buf_confirm, 160912d35f8SJens Axboe .release = page_cache_pipe_buf_release, 161912d35f8SJens Axboe .steal = user_page_pipe_buf_steal, 162f84d7519SJens Axboe .get = generic_pipe_buf_get, 163912d35f8SJens Axboe }; 164912d35f8SJens Axboe 165825cdcb1SNamhyung Kim static void wakeup_pipe_readers(struct pipe_inode_info *pipe) 166825cdcb1SNamhyung Kim { 167825cdcb1SNamhyung Kim smp_mb(); 168825cdcb1SNamhyung Kim if (waitqueue_active(&pipe->wait)) 169825cdcb1SNamhyung Kim wake_up_interruptible(&pipe->wait); 170825cdcb1SNamhyung Kim kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN); 171825cdcb1SNamhyung Kim } 172825cdcb1SNamhyung Kim 173932cc6d4SJens Axboe /** 174932cc6d4SJens Axboe * splice_to_pipe - fill passed data into a pipe 175932cc6d4SJens Axboe * @pipe: pipe to fill 176932cc6d4SJens Axboe * @spd: data to fill 177932cc6d4SJens Axboe * 178932cc6d4SJens Axboe * Description: 17979685b8dSRandy Dunlap * @spd contains a map of pages and len/offset tuples, along with 180932cc6d4SJens Axboe * the struct pipe_buf_operations associated with these pages. This 181932cc6d4SJens Axboe * function will link that data to the pipe. 182932cc6d4SJens Axboe * 18383f9135bSJens Axboe */ 184d6b29d7cSJens Axboe ssize_t splice_to_pipe(struct pipe_inode_info *pipe, 185912d35f8SJens Axboe struct splice_pipe_desc *spd) 1865274f052SJens Axboe { 18700de00bdSJens Axboe unsigned int spd_pages = spd->nr_pages; 1888924feffSAl Viro int ret = 0, page_nr = 0; 1895274f052SJens Axboe 190d6785d91SRabin Vincent if (!spd_pages) 191d6785d91SRabin Vincent return 0; 192d6785d91SRabin Vincent 1938924feffSAl Viro if (unlikely(!pipe->readers)) { 1945274f052SJens Axboe send_sig(SIGPIPE, current, 0); 1955274f052SJens Axboe ret = -EPIPE; 1968924feffSAl Viro goto out; 1975274f052SJens Axboe } 1985274f052SJens Axboe 1998924feffSAl Viro while (pipe->nrbufs < pipe->buffers) { 20035f3d14dSJens Axboe int newbuf = (pipe->curbuf + pipe->nrbufs) & (pipe->buffers - 1); 2013a326a2cSIngo Molnar struct pipe_buffer *buf = pipe->bufs + newbuf; 2025274f052SJens Axboe 203912d35f8SJens Axboe buf->page = spd->pages[page_nr]; 204912d35f8SJens Axboe buf->offset = spd->partial[page_nr].offset; 205912d35f8SJens Axboe buf->len = spd->partial[page_nr].len; 206497f9625SJens Axboe buf->private = spd->partial[page_nr].private; 207912d35f8SJens Axboe buf->ops = spd->ops; 2085a81e6a1SMiklos Szeredi buf->flags = 0; 2097afa6fd0SJens Axboe 2106f767b04SJens Axboe pipe->nrbufs++; 211912d35f8SJens Axboe page_nr++; 212912d35f8SJens Axboe ret += buf->len; 213912d35f8SJens Axboe 214912d35f8SJens Axboe if (!--spd->nr_pages) 2155274f052SJens Axboe break; 2165274f052SJens Axboe } 2175274f052SJens Axboe 21829e35094SLinus Torvalds if (!ret) 21929e35094SLinus Torvalds ret = -EAGAIN; 22029e35094SLinus Torvalds 2218924feffSAl Viro out: 22200de00bdSJens Axboe while (page_nr < spd_pages) 223bbdfc2f7SJens Axboe spd->spd_release(spd, page_nr++); 2245274f052SJens Axboe 2255274f052SJens Axboe return ret; 2265274f052SJens Axboe } 2272b514574SHannes Frederic Sowa EXPORT_SYMBOL_GPL(splice_to_pipe); 2285274f052SJens Axboe 22979fddc4eSAl Viro ssize_t add_to_pipe(struct pipe_inode_info *pipe, struct pipe_buffer *buf) 23079fddc4eSAl Viro { 23179fddc4eSAl Viro int ret; 23279fddc4eSAl Viro 23379fddc4eSAl Viro if (unlikely(!pipe->readers)) { 23479fddc4eSAl Viro send_sig(SIGPIPE, current, 0); 23579fddc4eSAl Viro ret = -EPIPE; 23679fddc4eSAl Viro } else if (pipe->nrbufs == pipe->buffers) { 23779fddc4eSAl Viro ret = -EAGAIN; 23879fddc4eSAl Viro } else { 23979fddc4eSAl Viro int newbuf = (pipe->curbuf + pipe->nrbufs) & (pipe->buffers - 1); 24079fddc4eSAl Viro pipe->bufs[newbuf] = *buf; 24179fddc4eSAl Viro pipe->nrbufs++; 24279fddc4eSAl Viro return buf->len; 24379fddc4eSAl Viro } 244a779638cSMiklos Szeredi pipe_buf_release(pipe, buf); 24579fddc4eSAl Viro return ret; 24679fddc4eSAl Viro } 24779fddc4eSAl Viro EXPORT_SYMBOL(add_to_pipe); 24879fddc4eSAl Viro 24935f3d14dSJens Axboe /* 25035f3d14dSJens Axboe * Check if we need to grow the arrays holding pages and partial page 25135f3d14dSJens Axboe * descriptions. 25235f3d14dSJens Axboe */ 253047fe360SEric Dumazet int splice_grow_spd(const struct pipe_inode_info *pipe, struct splice_pipe_desc *spd) 25435f3d14dSJens Axboe { 2556aa7de05SMark Rutland unsigned int buffers = READ_ONCE(pipe->buffers); 256047fe360SEric Dumazet 257047fe360SEric Dumazet spd->nr_pages_max = buffers; 258047fe360SEric Dumazet if (buffers <= PIPE_DEF_BUFFERS) 25935f3d14dSJens Axboe return 0; 26035f3d14dSJens Axboe 2616da2ec56SKees Cook spd->pages = kmalloc_array(buffers, sizeof(struct page *), GFP_KERNEL); 2626da2ec56SKees Cook spd->partial = kmalloc_array(buffers, sizeof(struct partial_page), 2636da2ec56SKees Cook GFP_KERNEL); 26435f3d14dSJens Axboe 26535f3d14dSJens Axboe if (spd->pages && spd->partial) 26635f3d14dSJens Axboe return 0; 26735f3d14dSJens Axboe 26835f3d14dSJens Axboe kfree(spd->pages); 26935f3d14dSJens Axboe kfree(spd->partial); 27035f3d14dSJens Axboe return -ENOMEM; 27135f3d14dSJens Axboe } 27235f3d14dSJens Axboe 273047fe360SEric Dumazet void splice_shrink_spd(struct splice_pipe_desc *spd) 27435f3d14dSJens Axboe { 275047fe360SEric Dumazet if (spd->nr_pages_max <= PIPE_DEF_BUFFERS) 27635f3d14dSJens Axboe return; 27735f3d14dSJens Axboe 27835f3d14dSJens Axboe kfree(spd->pages); 27935f3d14dSJens Axboe kfree(spd->partial); 28035f3d14dSJens Axboe } 28135f3d14dSJens Axboe 28283f9135bSJens Axboe /** 28383f9135bSJens Axboe * generic_file_splice_read - splice data from file to a pipe 28483f9135bSJens Axboe * @in: file to splice from 285932cc6d4SJens Axboe * @ppos: position in @in 28683f9135bSJens Axboe * @pipe: pipe to splice to 28783f9135bSJens Axboe * @len: number of bytes to splice 28883f9135bSJens Axboe * @flags: splice modifier flags 28983f9135bSJens Axboe * 290932cc6d4SJens Axboe * Description: 291932cc6d4SJens Axboe * Will read pages from given file and fill them into a pipe. Can be 29282c156f8SAl Viro * used as long as it has more or less sane ->read_iter(). 293932cc6d4SJens Axboe * 29483f9135bSJens Axboe */ 295cbb7e577SJens Axboe ssize_t generic_file_splice_read(struct file *in, loff_t *ppos, 296cbb7e577SJens Axboe struct pipe_inode_info *pipe, size_t len, 297cbb7e577SJens Axboe unsigned int flags) 2985274f052SJens Axboe { 29982c156f8SAl Viro struct iov_iter to; 30082c156f8SAl Viro struct kiocb kiocb; 30182c156f8SAl Viro int idx, ret; 302be64f884SBoaz Harrosh 303aa563d7bSDavid Howells iov_iter_pipe(&to, READ, pipe, len); 30482c156f8SAl Viro idx = to.idx; 30582c156f8SAl Viro init_sync_kiocb(&kiocb, in); 30682c156f8SAl Viro kiocb.ki_pos = *ppos; 307bb7462b6SMiklos Szeredi ret = call_read_iter(in, &kiocb, &to); 308723590edSMiklos Szeredi if (ret > 0) { 30982c156f8SAl Viro *ppos = kiocb.ki_pos; 310723590edSMiklos Szeredi file_accessed(in); 31182c156f8SAl Viro } else if (ret < 0) { 31282c156f8SAl Viro to.idx = idx; 31382c156f8SAl Viro to.iov_offset = 0; 31482c156f8SAl Viro iov_iter_advance(&to, 0); /* to free what was emitted */ 31582c156f8SAl Viro /* 31682c156f8SAl Viro * callers of ->splice_read() expect -EAGAIN on 31782c156f8SAl Viro * "can't put anything in there", rather than -EFAULT. 31882c156f8SAl Viro */ 31982c156f8SAl Viro if (ret == -EFAULT) 32082c156f8SAl Viro ret = -EAGAIN; 321723590edSMiklos Szeredi } 3225274f052SJens Axboe 3235274f052SJens Axboe return ret; 3245274f052SJens Axboe } 325059a8f37SJens Axboe EXPORT_SYMBOL(generic_file_splice_read); 326059a8f37SJens Axboe 327241699cdSAl Viro const struct pipe_buf_operations default_pipe_buf_ops = { 3286818173bSMiklos Szeredi .confirm = generic_pipe_buf_confirm, 3296818173bSMiklos Szeredi .release = generic_pipe_buf_release, 3306818173bSMiklos Szeredi .steal = generic_pipe_buf_steal, 3316818173bSMiklos Szeredi .get = generic_pipe_buf_get, 3326818173bSMiklos Szeredi }; 3336818173bSMiklos Szeredi 334b9872226SJann Horn int generic_pipe_buf_nosteal(struct pipe_inode_info *pipe, 33528a625cbSMiklos Szeredi struct pipe_buffer *buf) 33628a625cbSMiklos Szeredi { 33728a625cbSMiklos Szeredi return 1; 33828a625cbSMiklos Szeredi } 33928a625cbSMiklos Szeredi 34028a625cbSMiklos Szeredi /* Pipe buffer operations for a socket and similar. */ 34128a625cbSMiklos Szeredi const struct pipe_buf_operations nosteal_pipe_buf_ops = { 34228a625cbSMiklos Szeredi .confirm = generic_pipe_buf_confirm, 34328a625cbSMiklos Szeredi .release = generic_pipe_buf_release, 34428a625cbSMiklos Szeredi .steal = generic_pipe_buf_nosteal, 34528a625cbSMiklos Szeredi .get = generic_pipe_buf_get, 34628a625cbSMiklos Szeredi }; 34728a625cbSMiklos Szeredi EXPORT_SYMBOL(nosteal_pipe_buf_ops); 34828a625cbSMiklos Szeredi 349523ac9afSAl Viro static ssize_t kernel_readv(struct file *file, const struct kvec *vec, 3506818173bSMiklos Szeredi unsigned long vlen, loff_t offset) 3516818173bSMiklos Szeredi { 3526818173bSMiklos Szeredi mm_segment_t old_fs; 3536818173bSMiklos Szeredi loff_t pos = offset; 3546818173bSMiklos Szeredi ssize_t res; 3556818173bSMiklos Szeredi 3566818173bSMiklos Szeredi old_fs = get_fs(); 357736706beSLinus Torvalds set_fs(KERNEL_DS); 3586818173bSMiklos Szeredi /* The cast to a user pointer is valid due to the set_fs() */ 359793b80efSChristoph Hellwig res = vfs_readv(file, (const struct iovec __user *)vec, vlen, &pos, 0); 3606818173bSMiklos Szeredi set_fs(old_fs); 3616818173bSMiklos Szeredi 3626818173bSMiklos Szeredi return res; 3636818173bSMiklos Szeredi } 3646818173bSMiklos Szeredi 36582c156f8SAl Viro static ssize_t default_file_splice_read(struct file *in, loff_t *ppos, 3666818173bSMiklos Szeredi struct pipe_inode_info *pipe, size_t len, 3676818173bSMiklos Szeredi unsigned int flags) 3686818173bSMiklos Szeredi { 369523ac9afSAl Viro struct kvec *vec, __vec[PIPE_DEF_BUFFERS]; 370523ac9afSAl Viro struct iov_iter to; 371523ac9afSAl Viro struct page **pages; 3726818173bSMiklos Szeredi unsigned int nr_pages; 37313c0f52bSAl Viro size_t offset, base, copied = 0; 3746818173bSMiklos Szeredi ssize_t res; 3756818173bSMiklos Szeredi int i; 3766818173bSMiklos Szeredi 377523ac9afSAl Viro if (pipe->nrbufs == pipe->buffers) 378523ac9afSAl Viro return -EAGAIN; 379523ac9afSAl Viro 380523ac9afSAl Viro /* 381523ac9afSAl Viro * Try to keep page boundaries matching to source pagecache ones - 382523ac9afSAl Viro * it probably won't be much help, but... 383523ac9afSAl Viro */ 384523ac9afSAl Viro offset = *ppos & ~PAGE_MASK; 385523ac9afSAl Viro 386aa563d7bSDavid Howells iov_iter_pipe(&to, READ, pipe, len + offset); 387523ac9afSAl Viro 38813c0f52bSAl Viro res = iov_iter_get_pages_alloc(&to, &pages, len + offset, &base); 389523ac9afSAl Viro if (res <= 0) 39035f3d14dSJens Axboe return -ENOMEM; 39135f3d14dSJens Axboe 39213c0f52bSAl Viro nr_pages = DIV_ROUND_UP(res + base, PAGE_SIZE); 393523ac9afSAl Viro 39435f3d14dSJens Axboe vec = __vec; 395523ac9afSAl Viro if (nr_pages > PIPE_DEF_BUFFERS) { 3966da2ec56SKees Cook vec = kmalloc_array(nr_pages, sizeof(struct kvec), GFP_KERNEL); 397523ac9afSAl Viro if (unlikely(!vec)) { 398523ac9afSAl Viro res = -ENOMEM; 399523ac9afSAl Viro goto out; 400523ac9afSAl Viro } 40135f3d14dSJens Axboe } 40235f3d14dSJens Axboe 403523ac9afSAl Viro pipe->bufs[to.idx].offset = offset; 404523ac9afSAl Viro pipe->bufs[to.idx].len -= offset; 4056818173bSMiklos Szeredi 406523ac9afSAl Viro for (i = 0; i < nr_pages; i++) { 407523ac9afSAl Viro size_t this_len = min_t(size_t, len, PAGE_SIZE - offset); 408523ac9afSAl Viro vec[i].iov_base = page_address(pages[i]) + offset; 4096818173bSMiklos Szeredi vec[i].iov_len = this_len; 4106818173bSMiklos Szeredi len -= this_len; 4116818173bSMiklos Szeredi offset = 0; 4126818173bSMiklos Szeredi } 4136818173bSMiklos Szeredi 414523ac9afSAl Viro res = kernel_readv(in, vec, nr_pages, *ppos); 415523ac9afSAl Viro if (res > 0) { 416523ac9afSAl Viro copied = res; 4176818173bSMiklos Szeredi *ppos += res; 418523ac9afSAl Viro } 4196818173bSMiklos Szeredi 42035f3d14dSJens Axboe if (vec != __vec) 42135f3d14dSJens Axboe kfree(vec); 422523ac9afSAl Viro out: 423523ac9afSAl Viro for (i = 0; i < nr_pages; i++) 424523ac9afSAl Viro put_page(pages[i]); 425523ac9afSAl Viro kvfree(pages); 426523ac9afSAl Viro iov_iter_advance(&to, copied); /* truncates and discards */ 4276818173bSMiklos Szeredi return res; 4286818173bSMiklos Szeredi } 4296818173bSMiklos Szeredi 4305274f052SJens Axboe /* 4314f6f0bd2SJens Axboe * Send 'sd->len' bytes to socket from 'sd->file' at position 'sd->pos' 432016b661eSJens Axboe * using sendpage(). Return the number of bytes sent. 4335274f052SJens Axboe */ 43476ad4d11SJens Axboe static int pipe_to_sendpage(struct pipe_inode_info *pipe, 4355274f052SJens Axboe struct pipe_buffer *buf, struct splice_desc *sd) 4365274f052SJens Axboe { 4376a14b90bSJens Axboe struct file *file = sd->u.file; 4385274f052SJens Axboe loff_t pos = sd->pos; 439a8adbe37SMichał Mirosław int more; 4405274f052SJens Axboe 44172c2d531SAl Viro if (!likely(file->f_op->sendpage)) 442a8adbe37SMichał Mirosław return -EINVAL; 443a8adbe37SMichał Mirosław 44435f9c09fSEric Dumazet more = (sd->flags & SPLICE_F_MORE) ? MSG_MORE : 0; 445ae62ca7bSEric Dumazet 446ae62ca7bSEric Dumazet if (sd->len < sd->total_len && pipe->nrbufs > 1) 44735f9c09fSEric Dumazet more |= MSG_SENDPAGE_NOTLAST; 448ae62ca7bSEric Dumazet 449a8adbe37SMichał Mirosław return file->f_op->sendpage(file, buf->page, buf->offset, 450f84d7519SJens Axboe sd->len, &pos, more); 4515274f052SJens Axboe } 4525274f052SJens Axboe 453b3c2d2ddSMiklos Szeredi static void wakeup_pipe_writers(struct pipe_inode_info *pipe) 454b3c2d2ddSMiklos Szeredi { 455b3c2d2ddSMiklos Szeredi smp_mb(); 456b3c2d2ddSMiklos Szeredi if (waitqueue_active(&pipe->wait)) 457b3c2d2ddSMiklos Szeredi wake_up_interruptible(&pipe->wait); 458b3c2d2ddSMiklos Szeredi kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT); 459b3c2d2ddSMiklos Szeredi } 460b3c2d2ddSMiklos Szeredi 461b3c2d2ddSMiklos Szeredi /** 462b3c2d2ddSMiklos Szeredi * splice_from_pipe_feed - feed available data from a pipe to a file 463b3c2d2ddSMiklos Szeredi * @pipe: pipe to splice from 464b3c2d2ddSMiklos Szeredi * @sd: information to @actor 465b3c2d2ddSMiklos Szeredi * @actor: handler that splices the data 466b3c2d2ddSMiklos Szeredi * 467b3c2d2ddSMiklos Szeredi * Description: 468b3c2d2ddSMiklos Szeredi * This function loops over the pipe and calls @actor to do the 469b3c2d2ddSMiklos Szeredi * actual moving of a single struct pipe_buffer to the desired 470b3c2d2ddSMiklos Szeredi * destination. It returns when there's no more buffers left in 471b3c2d2ddSMiklos Szeredi * the pipe or if the requested number of bytes (@sd->total_len) 472b3c2d2ddSMiklos Szeredi * have been copied. It returns a positive number (one) if the 473b3c2d2ddSMiklos Szeredi * pipe needs to be filled with more data, zero if the required 474b3c2d2ddSMiklos Szeredi * number of bytes have been copied and -errno on error. 475b3c2d2ddSMiklos Szeredi * 476b3c2d2ddSMiklos Szeredi * This, together with splice_from_pipe_{begin,end,next}, may be 477b3c2d2ddSMiklos Szeredi * used to implement the functionality of __splice_from_pipe() when 478b3c2d2ddSMiklos Szeredi * locking is required around copying the pipe buffers to the 479b3c2d2ddSMiklos Szeredi * destination. 480b3c2d2ddSMiklos Szeredi */ 48196f9bc8fSAl Viro static int splice_from_pipe_feed(struct pipe_inode_info *pipe, struct splice_desc *sd, 482b3c2d2ddSMiklos Szeredi splice_actor *actor) 483b3c2d2ddSMiklos Szeredi { 484b3c2d2ddSMiklos Szeredi int ret; 485b3c2d2ddSMiklos Szeredi 486b3c2d2ddSMiklos Szeredi while (pipe->nrbufs) { 487b3c2d2ddSMiklos Szeredi struct pipe_buffer *buf = pipe->bufs + pipe->curbuf; 488b3c2d2ddSMiklos Szeredi 489b3c2d2ddSMiklos Szeredi sd->len = buf->len; 490b3c2d2ddSMiklos Szeredi if (sd->len > sd->total_len) 491b3c2d2ddSMiklos Szeredi sd->len = sd->total_len; 492b3c2d2ddSMiklos Szeredi 493fba597dbSMiklos Szeredi ret = pipe_buf_confirm(pipe, buf); 494a8adbe37SMichał Mirosław if (unlikely(ret)) { 495b3c2d2ddSMiklos Szeredi if (ret == -ENODATA) 496b3c2d2ddSMiklos Szeredi ret = 0; 497b3c2d2ddSMiklos Szeredi return ret; 498b3c2d2ddSMiklos Szeredi } 499a8adbe37SMichał Mirosław 500a8adbe37SMichał Mirosław ret = actor(pipe, buf, sd); 501a8adbe37SMichał Mirosław if (ret <= 0) 502a8adbe37SMichał Mirosław return ret; 503a8adbe37SMichał Mirosław 504b3c2d2ddSMiklos Szeredi buf->offset += ret; 505b3c2d2ddSMiklos Szeredi buf->len -= ret; 506b3c2d2ddSMiklos Szeredi 507b3c2d2ddSMiklos Szeredi sd->num_spliced += ret; 508b3c2d2ddSMiklos Szeredi sd->len -= ret; 509b3c2d2ddSMiklos Szeredi sd->pos += ret; 510b3c2d2ddSMiklos Szeredi sd->total_len -= ret; 511b3c2d2ddSMiklos Szeredi 512b3c2d2ddSMiklos Szeredi if (!buf->len) { 513a779638cSMiklos Szeredi pipe_buf_release(pipe, buf); 51435f3d14dSJens Axboe pipe->curbuf = (pipe->curbuf + 1) & (pipe->buffers - 1); 515b3c2d2ddSMiklos Szeredi pipe->nrbufs--; 5166447a3cfSAl Viro if (pipe->files) 517b3c2d2ddSMiklos Szeredi sd->need_wakeup = true; 518b3c2d2ddSMiklos Szeredi } 519b3c2d2ddSMiklos Szeredi 520b3c2d2ddSMiklos Szeredi if (!sd->total_len) 521b3c2d2ddSMiklos Szeredi return 0; 522b3c2d2ddSMiklos Szeredi } 523b3c2d2ddSMiklos Szeredi 524b3c2d2ddSMiklos Szeredi return 1; 525b3c2d2ddSMiklos Szeredi } 526b3c2d2ddSMiklos Szeredi 527b3c2d2ddSMiklos Szeredi /** 528b3c2d2ddSMiklos Szeredi * splice_from_pipe_next - wait for some data to splice from 529b3c2d2ddSMiklos Szeredi * @pipe: pipe to splice from 530b3c2d2ddSMiklos Szeredi * @sd: information about the splice operation 531b3c2d2ddSMiklos Szeredi * 532b3c2d2ddSMiklos Szeredi * Description: 533b3c2d2ddSMiklos Szeredi * This function will wait for some data and return a positive 534b3c2d2ddSMiklos Szeredi * value (one) if pipe buffers are available. It will return zero 535b3c2d2ddSMiklos Szeredi * or -errno if no more data needs to be spliced. 536b3c2d2ddSMiklos Szeredi */ 53796f9bc8fSAl Viro static int splice_from_pipe_next(struct pipe_inode_info *pipe, struct splice_desc *sd) 538b3c2d2ddSMiklos Szeredi { 539c725bfceSJan Kara /* 540c725bfceSJan Kara * Check for signal early to make process killable when there are 541c725bfceSJan Kara * always buffers available 542c725bfceSJan Kara */ 543c725bfceSJan Kara if (signal_pending(current)) 544c725bfceSJan Kara return -ERESTARTSYS; 545c725bfceSJan Kara 546b3c2d2ddSMiklos Szeredi while (!pipe->nrbufs) { 547b3c2d2ddSMiklos Szeredi if (!pipe->writers) 548b3c2d2ddSMiklos Szeredi return 0; 549b3c2d2ddSMiklos Szeredi 550b3c2d2ddSMiklos Szeredi if (!pipe->waiting_writers && sd->num_spliced) 551b3c2d2ddSMiklos Szeredi return 0; 552b3c2d2ddSMiklos Szeredi 553b3c2d2ddSMiklos Szeredi if (sd->flags & SPLICE_F_NONBLOCK) 554b3c2d2ddSMiklos Szeredi return -EAGAIN; 555b3c2d2ddSMiklos Szeredi 556b3c2d2ddSMiklos Szeredi if (signal_pending(current)) 557b3c2d2ddSMiklos Szeredi return -ERESTARTSYS; 558b3c2d2ddSMiklos Szeredi 559b3c2d2ddSMiklos Szeredi if (sd->need_wakeup) { 560b3c2d2ddSMiklos Szeredi wakeup_pipe_writers(pipe); 561b3c2d2ddSMiklos Szeredi sd->need_wakeup = false; 562b3c2d2ddSMiklos Szeredi } 563b3c2d2ddSMiklos Szeredi 564b3c2d2ddSMiklos Szeredi pipe_wait(pipe); 565b3c2d2ddSMiklos Szeredi } 566b3c2d2ddSMiklos Szeredi 567b3c2d2ddSMiklos Szeredi return 1; 568b3c2d2ddSMiklos Szeredi } 569b3c2d2ddSMiklos Szeredi 570b3c2d2ddSMiklos Szeredi /** 571b3c2d2ddSMiklos Szeredi * splice_from_pipe_begin - start splicing from pipe 572b80901bbSRandy Dunlap * @sd: information about the splice operation 573b3c2d2ddSMiklos Szeredi * 574b3c2d2ddSMiklos Szeredi * Description: 575b3c2d2ddSMiklos Szeredi * This function should be called before a loop containing 576b3c2d2ddSMiklos Szeredi * splice_from_pipe_next() and splice_from_pipe_feed() to 577b3c2d2ddSMiklos Szeredi * initialize the necessary fields of @sd. 578b3c2d2ddSMiklos Szeredi */ 57996f9bc8fSAl Viro static void splice_from_pipe_begin(struct splice_desc *sd) 580b3c2d2ddSMiklos Szeredi { 581b3c2d2ddSMiklos Szeredi sd->num_spliced = 0; 582b3c2d2ddSMiklos Szeredi sd->need_wakeup = false; 583b3c2d2ddSMiklos Szeredi } 584b3c2d2ddSMiklos Szeredi 585b3c2d2ddSMiklos Szeredi /** 586b3c2d2ddSMiklos Szeredi * splice_from_pipe_end - finish splicing from pipe 587b3c2d2ddSMiklos Szeredi * @pipe: pipe to splice from 588b3c2d2ddSMiklos Szeredi * @sd: information about the splice operation 589b3c2d2ddSMiklos Szeredi * 590b3c2d2ddSMiklos Szeredi * Description: 591b3c2d2ddSMiklos Szeredi * This function will wake up pipe writers if necessary. It should 592b3c2d2ddSMiklos Szeredi * be called after a loop containing splice_from_pipe_next() and 593b3c2d2ddSMiklos Szeredi * splice_from_pipe_feed(). 594b3c2d2ddSMiklos Szeredi */ 59596f9bc8fSAl Viro static void splice_from_pipe_end(struct pipe_inode_info *pipe, struct splice_desc *sd) 596b3c2d2ddSMiklos Szeredi { 597b3c2d2ddSMiklos Szeredi if (sd->need_wakeup) 598b3c2d2ddSMiklos Szeredi wakeup_pipe_writers(pipe); 599b3c2d2ddSMiklos Szeredi } 600b3c2d2ddSMiklos Szeredi 601932cc6d4SJens Axboe /** 602932cc6d4SJens Axboe * __splice_from_pipe - splice data from a pipe to given actor 603932cc6d4SJens Axboe * @pipe: pipe to splice from 604932cc6d4SJens Axboe * @sd: information to @actor 605932cc6d4SJens Axboe * @actor: handler that splices the data 606932cc6d4SJens Axboe * 607932cc6d4SJens Axboe * Description: 608932cc6d4SJens Axboe * This function does little more than loop over the pipe and call 609932cc6d4SJens Axboe * @actor to do the actual moving of a single struct pipe_buffer to 610932cc6d4SJens Axboe * the desired destination. See pipe_to_file, pipe_to_sendpage, or 611932cc6d4SJens Axboe * pipe_to_user. 612932cc6d4SJens Axboe * 61383f9135bSJens Axboe */ 614c66ab6faSJens Axboe ssize_t __splice_from_pipe(struct pipe_inode_info *pipe, struct splice_desc *sd, 615c66ab6faSJens Axboe splice_actor *actor) 6165274f052SJens Axboe { 617b3c2d2ddSMiklos Szeredi int ret; 6185274f052SJens Axboe 619b3c2d2ddSMiklos Szeredi splice_from_pipe_begin(sd); 620b3c2d2ddSMiklos Szeredi do { 621c2489e07SJan Kara cond_resched(); 622b3c2d2ddSMiklos Szeredi ret = splice_from_pipe_next(pipe, sd); 623b3c2d2ddSMiklos Szeredi if (ret > 0) 624b3c2d2ddSMiklos Szeredi ret = splice_from_pipe_feed(pipe, sd, actor); 625b3c2d2ddSMiklos Szeredi } while (ret > 0); 626b3c2d2ddSMiklos Szeredi splice_from_pipe_end(pipe, sd); 6275274f052SJens Axboe 628b3c2d2ddSMiklos Szeredi return sd->num_spliced ? sd->num_spliced : ret; 6295274f052SJens Axboe } 63040bee44eSMark Fasheh EXPORT_SYMBOL(__splice_from_pipe); 6315274f052SJens Axboe 632932cc6d4SJens Axboe /** 633932cc6d4SJens Axboe * splice_from_pipe - splice data from a pipe to a file 634932cc6d4SJens Axboe * @pipe: pipe to splice from 635932cc6d4SJens Axboe * @out: file to splice to 636932cc6d4SJens Axboe * @ppos: position in @out 637932cc6d4SJens Axboe * @len: how many bytes to splice 638932cc6d4SJens Axboe * @flags: splice modifier flags 639932cc6d4SJens Axboe * @actor: handler that splices the data 640932cc6d4SJens Axboe * 641932cc6d4SJens Axboe * Description: 6422933970bSMiklos Szeredi * See __splice_from_pipe. This function locks the pipe inode, 643932cc6d4SJens Axboe * otherwise it's identical to __splice_from_pipe(). 644932cc6d4SJens Axboe * 645932cc6d4SJens Axboe */ 6466da61809SMark Fasheh ssize_t splice_from_pipe(struct pipe_inode_info *pipe, struct file *out, 6476da61809SMark Fasheh loff_t *ppos, size_t len, unsigned int flags, 6486da61809SMark Fasheh splice_actor *actor) 6496da61809SMark Fasheh { 6506da61809SMark Fasheh ssize_t ret; 651c66ab6faSJens Axboe struct splice_desc sd = { 652c66ab6faSJens Axboe .total_len = len, 653c66ab6faSJens Axboe .flags = flags, 654c66ab6faSJens Axboe .pos = *ppos, 6556a14b90bSJens Axboe .u.file = out, 656c66ab6faSJens Axboe }; 6576da61809SMark Fasheh 65861e0d47cSMiklos Szeredi pipe_lock(pipe); 659c66ab6faSJens Axboe ret = __splice_from_pipe(pipe, &sd, actor); 66061e0d47cSMiklos Szeredi pipe_unlock(pipe); 6616da61809SMark Fasheh 6626da61809SMark Fasheh return ret; 6636da61809SMark Fasheh } 6646da61809SMark Fasheh 6656da61809SMark Fasheh /** 6668d020765SAl Viro * iter_file_splice_write - splice data from a pipe to a file 6678d020765SAl Viro * @pipe: pipe info 6688d020765SAl Viro * @out: file to write to 6698d020765SAl Viro * @ppos: position in @out 6708d020765SAl Viro * @len: number of bytes to splice 6718d020765SAl Viro * @flags: splice modifier flags 6728d020765SAl Viro * 6738d020765SAl Viro * Description: 6748d020765SAl Viro * Will either move or copy pages (determined by @flags options) from 6758d020765SAl Viro * the given pipe inode to the given file. 6768d020765SAl Viro * This one is ->write_iter-based. 6778d020765SAl Viro * 6788d020765SAl Viro */ 6798d020765SAl Viro ssize_t 6808d020765SAl Viro iter_file_splice_write(struct pipe_inode_info *pipe, struct file *out, 6818d020765SAl Viro loff_t *ppos, size_t len, unsigned int flags) 6828d020765SAl Viro { 6838d020765SAl Viro struct splice_desc sd = { 6848d020765SAl Viro .total_len = len, 6858d020765SAl Viro .flags = flags, 6868d020765SAl Viro .pos = *ppos, 6878d020765SAl Viro .u.file = out, 6888d020765SAl Viro }; 6898d020765SAl Viro int nbufs = pipe->buffers; 6908d020765SAl Viro struct bio_vec *array = kcalloc(nbufs, sizeof(struct bio_vec), 6918d020765SAl Viro GFP_KERNEL); 6928d020765SAl Viro ssize_t ret; 6938d020765SAl Viro 6948d020765SAl Viro if (unlikely(!array)) 6958d020765SAl Viro return -ENOMEM; 6968d020765SAl Viro 6978d020765SAl Viro pipe_lock(pipe); 6988d020765SAl Viro 6998d020765SAl Viro splice_from_pipe_begin(&sd); 7008d020765SAl Viro while (sd.total_len) { 7018d020765SAl Viro struct iov_iter from; 7028d020765SAl Viro size_t left; 7038d020765SAl Viro int n, idx; 7048d020765SAl Viro 7058d020765SAl Viro ret = splice_from_pipe_next(pipe, &sd); 7068d020765SAl Viro if (ret <= 0) 7078d020765SAl Viro break; 7088d020765SAl Viro 7098d020765SAl Viro if (unlikely(nbufs < pipe->buffers)) { 7108d020765SAl Viro kfree(array); 7118d020765SAl Viro nbufs = pipe->buffers; 7128d020765SAl Viro array = kcalloc(nbufs, sizeof(struct bio_vec), 7138d020765SAl Viro GFP_KERNEL); 7148d020765SAl Viro if (!array) { 7158d020765SAl Viro ret = -ENOMEM; 7168d020765SAl Viro break; 7178d020765SAl Viro } 7188d020765SAl Viro } 7198d020765SAl Viro 7208d020765SAl Viro /* build the vector */ 7218d020765SAl Viro left = sd.total_len; 7228d020765SAl Viro for (n = 0, idx = pipe->curbuf; left && n < pipe->nrbufs; n++, idx++) { 7238d020765SAl Viro struct pipe_buffer *buf = pipe->bufs + idx; 7248d020765SAl Viro size_t this_len = buf->len; 7258d020765SAl Viro 7268d020765SAl Viro if (this_len > left) 7278d020765SAl Viro this_len = left; 7288d020765SAl Viro 7298d020765SAl Viro if (idx == pipe->buffers - 1) 7308d020765SAl Viro idx = -1; 7318d020765SAl Viro 732fba597dbSMiklos Szeredi ret = pipe_buf_confirm(pipe, buf); 7338d020765SAl Viro if (unlikely(ret)) { 7348d020765SAl Viro if (ret == -ENODATA) 7358d020765SAl Viro ret = 0; 7368d020765SAl Viro goto done; 7378d020765SAl Viro } 7388d020765SAl Viro 7398d020765SAl Viro array[n].bv_page = buf->page; 7408d020765SAl Viro array[n].bv_len = this_len; 7418d020765SAl Viro array[n].bv_offset = buf->offset; 7428d020765SAl Viro left -= this_len; 7438d020765SAl Viro } 7448d020765SAl Viro 745aa563d7bSDavid Howells iov_iter_bvec(&from, WRITE, array, n, sd.total_len - left); 746abbb6589SChristoph Hellwig ret = vfs_iter_write(out, &from, &sd.pos, 0); 7478d020765SAl Viro if (ret <= 0) 7488d020765SAl Viro break; 7498d020765SAl Viro 7508d020765SAl Viro sd.num_spliced += ret; 7518d020765SAl Viro sd.total_len -= ret; 752dbe4e192SChristoph Hellwig *ppos = sd.pos; 7538d020765SAl Viro 7548d020765SAl Viro /* dismiss the fully eaten buffers, adjust the partial one */ 7558d020765SAl Viro while (ret) { 7568d020765SAl Viro struct pipe_buffer *buf = pipe->bufs + pipe->curbuf; 7578d020765SAl Viro if (ret >= buf->len) { 7588d020765SAl Viro ret -= buf->len; 7598d020765SAl Viro buf->len = 0; 760a779638cSMiklos Szeredi pipe_buf_release(pipe, buf); 7618d020765SAl Viro pipe->curbuf = (pipe->curbuf + 1) & (pipe->buffers - 1); 7628d020765SAl Viro pipe->nrbufs--; 7638d020765SAl Viro if (pipe->files) 7648d020765SAl Viro sd.need_wakeup = true; 7658d020765SAl Viro } else { 7668d020765SAl Viro buf->offset += ret; 7678d020765SAl Viro buf->len -= ret; 7688d020765SAl Viro ret = 0; 7698d020765SAl Viro } 7708d020765SAl Viro } 7718d020765SAl Viro } 7728d020765SAl Viro done: 7738d020765SAl Viro kfree(array); 7748d020765SAl Viro splice_from_pipe_end(pipe, &sd); 7758d020765SAl Viro 7768d020765SAl Viro pipe_unlock(pipe); 7778d020765SAl Viro 7788d020765SAl Viro if (sd.num_spliced) 7798d020765SAl Viro ret = sd.num_spliced; 7808d020765SAl Viro 7818d020765SAl Viro return ret; 7828d020765SAl Viro } 7838d020765SAl Viro 7848d020765SAl Viro EXPORT_SYMBOL(iter_file_splice_write); 7858d020765SAl Viro 786b2858d7dSMiklos Szeredi static int write_pipe_buf(struct pipe_inode_info *pipe, struct pipe_buffer *buf, 787b2858d7dSMiklos Szeredi struct splice_desc *sd) 7880b0a47f5SMiklos Szeredi { 789b2858d7dSMiklos Szeredi int ret; 790b2858d7dSMiklos Szeredi void *data; 79106ae43f3SAl Viro loff_t tmp = sd->pos; 792b2858d7dSMiklos Szeredi 793fbb32750SAl Viro data = kmap(buf->page); 79406ae43f3SAl Viro ret = __kernel_write(sd->u.file, data + buf->offset, sd->len, &tmp); 795fbb32750SAl Viro kunmap(buf->page); 796b2858d7dSMiklos Szeredi 797b2858d7dSMiklos Szeredi return ret; 7980b0a47f5SMiklos Szeredi } 7990b0a47f5SMiklos Szeredi 8000b0a47f5SMiklos Szeredi static ssize_t default_file_splice_write(struct pipe_inode_info *pipe, 8010b0a47f5SMiklos Szeredi struct file *out, loff_t *ppos, 8020b0a47f5SMiklos Szeredi size_t len, unsigned int flags) 8030b0a47f5SMiklos Szeredi { 804b2858d7dSMiklos Szeredi ssize_t ret; 8050b0a47f5SMiklos Szeredi 806b2858d7dSMiklos Szeredi ret = splice_from_pipe(pipe, out, ppos, len, flags, write_pipe_buf); 807b2858d7dSMiklos Szeredi if (ret > 0) 808b2858d7dSMiklos Szeredi *ppos += ret; 8090b0a47f5SMiklos Szeredi 810b2858d7dSMiklos Szeredi return ret; 8110b0a47f5SMiklos Szeredi } 8120b0a47f5SMiklos Szeredi 81383f9135bSJens Axboe /** 81483f9135bSJens Axboe * generic_splice_sendpage - splice data from a pipe to a socket 815932cc6d4SJens Axboe * @pipe: pipe to splice from 81683f9135bSJens Axboe * @out: socket to write to 817932cc6d4SJens Axboe * @ppos: position in @out 81883f9135bSJens Axboe * @len: number of bytes to splice 81983f9135bSJens Axboe * @flags: splice modifier flags 82083f9135bSJens Axboe * 821932cc6d4SJens Axboe * Description: 82283f9135bSJens Axboe * Will send @len bytes from the pipe to a network socket. No data copying 82383f9135bSJens Axboe * is involved. 82483f9135bSJens Axboe * 82583f9135bSJens Axboe */ 8263a326a2cSIngo Molnar ssize_t generic_splice_sendpage(struct pipe_inode_info *pipe, struct file *out, 827cbb7e577SJens Axboe loff_t *ppos, size_t len, unsigned int flags) 8285274f052SJens Axboe { 82900522fb4SJens Axboe return splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_sendpage); 8305274f052SJens Axboe } 8315274f052SJens Axboe 832059a8f37SJens Axboe EXPORT_SYMBOL(generic_splice_sendpage); 833a0f06780SJeff Garzik 83483f9135bSJens Axboe /* 83583f9135bSJens Axboe * Attempt to initiate a splice from pipe to file. 83683f9135bSJens Axboe */ 8373a326a2cSIngo Molnar static long do_splice_from(struct pipe_inode_info *pipe, struct file *out, 838cbb7e577SJens Axboe loff_t *ppos, size_t len, unsigned int flags) 8395274f052SJens Axboe { 8400b0a47f5SMiklos Szeredi ssize_t (*splice_write)(struct pipe_inode_info *, struct file *, 8410b0a47f5SMiklos Szeredi loff_t *, size_t, unsigned int); 8425274f052SJens Axboe 84372c2d531SAl Viro if (out->f_op->splice_write) 8440b0a47f5SMiklos Szeredi splice_write = out->f_op->splice_write; 845cc56f7deSChangli Gao else 8460b0a47f5SMiklos Szeredi splice_write = default_file_splice_write; 8470b0a47f5SMiklos Szeredi 848500368f7SAl Viro return splice_write(pipe, out, ppos, len, flags); 8495274f052SJens Axboe } 8505274f052SJens Axboe 85183f9135bSJens Axboe /* 85283f9135bSJens Axboe * Attempt to initiate a splice from a file to a pipe. 85383f9135bSJens Axboe */ 854cbb7e577SJens Axboe static long do_splice_to(struct file *in, loff_t *ppos, 855cbb7e577SJens Axboe struct pipe_inode_info *pipe, size_t len, 856cbb7e577SJens Axboe unsigned int flags) 8575274f052SJens Axboe { 8586818173bSMiklos Szeredi ssize_t (*splice_read)(struct file *, loff_t *, 8596818173bSMiklos Szeredi struct pipe_inode_info *, size_t, unsigned int); 8605274f052SJens Axboe int ret; 8615274f052SJens Axboe 86249570e9bSJens Axboe if (unlikely(!(in->f_mode & FMODE_READ))) 8635274f052SJens Axboe return -EBADF; 8645274f052SJens Axboe 865cbb7e577SJens Axboe ret = rw_verify_area(READ, in, ppos, len); 8665274f052SJens Axboe if (unlikely(ret < 0)) 8675274f052SJens Axboe return ret; 8685274f052SJens Axboe 86903cc0789SAl Viro if (unlikely(len > MAX_RW_COUNT)) 87003cc0789SAl Viro len = MAX_RW_COUNT; 87103cc0789SAl Viro 87272c2d531SAl Viro if (in->f_op->splice_read) 8736818173bSMiklos Szeredi splice_read = in->f_op->splice_read; 874cc56f7deSChangli Gao else 8756818173bSMiklos Szeredi splice_read = default_file_splice_read; 8766818173bSMiklos Szeredi 8776818173bSMiklos Szeredi return splice_read(in, ppos, pipe, len, flags); 8785274f052SJens Axboe } 8795274f052SJens Axboe 880932cc6d4SJens Axboe /** 881932cc6d4SJens Axboe * splice_direct_to_actor - splices data directly between two non-pipes 882932cc6d4SJens Axboe * @in: file to splice from 883932cc6d4SJens Axboe * @sd: actor information on where to splice to 884932cc6d4SJens Axboe * @actor: handles the data splicing 885932cc6d4SJens Axboe * 886932cc6d4SJens Axboe * Description: 887932cc6d4SJens Axboe * This is a special case helper to splice directly between two 888932cc6d4SJens Axboe * points, without requiring an explicit pipe. Internally an allocated 889932cc6d4SJens Axboe * pipe is cached in the process, and reused during the lifetime of 890932cc6d4SJens Axboe * that process. 891932cc6d4SJens Axboe * 892c66ab6faSJens Axboe */ 893c66ab6faSJens Axboe ssize_t splice_direct_to_actor(struct file *in, struct splice_desc *sd, 894c66ab6faSJens Axboe splice_direct_actor *actor) 895b92ce558SJens Axboe { 896b92ce558SJens Axboe struct pipe_inode_info *pipe; 897b92ce558SJens Axboe long ret, bytes; 898b92ce558SJens Axboe umode_t i_mode; 899c66ab6faSJens Axboe size_t len; 9000ff28d9fSChristophe Leroy int i, flags, more; 901b92ce558SJens Axboe 902b92ce558SJens Axboe /* 903b92ce558SJens Axboe * We require the input being a regular file, as we don't want to 904b92ce558SJens Axboe * randomly drop data for eg socket -> socket splicing. Use the 905b92ce558SJens Axboe * piped splicing for that! 906b92ce558SJens Axboe */ 907496ad9aaSAl Viro i_mode = file_inode(in)->i_mode; 908b92ce558SJens Axboe if (unlikely(!S_ISREG(i_mode) && !S_ISBLK(i_mode))) 909b92ce558SJens Axboe return -EINVAL; 910b92ce558SJens Axboe 911b92ce558SJens Axboe /* 912b92ce558SJens Axboe * neither in nor out is a pipe, setup an internal pipe attached to 913b92ce558SJens Axboe * 'out' and transfer the wanted data from 'in' to 'out' through that 914b92ce558SJens Axboe */ 915b92ce558SJens Axboe pipe = current->splice_pipe; 91649570e9bSJens Axboe if (unlikely(!pipe)) { 9177bee130eSAl Viro pipe = alloc_pipe_info(); 918b92ce558SJens Axboe if (!pipe) 919b92ce558SJens Axboe return -ENOMEM; 920b92ce558SJens Axboe 921b92ce558SJens Axboe /* 922b92ce558SJens Axboe * We don't have an immediate reader, but we'll read the stuff 92300522fb4SJens Axboe * out of the pipe right after the splice_to_pipe(). So set 924b92ce558SJens Axboe * PIPE_READERS appropriately. 925b92ce558SJens Axboe */ 926b92ce558SJens Axboe pipe->readers = 1; 927b92ce558SJens Axboe 928b92ce558SJens Axboe current->splice_pipe = pipe; 929b92ce558SJens Axboe } 930b92ce558SJens Axboe 931b92ce558SJens Axboe /* 93273d62d83SIngo Molnar * Do the splice. 933b92ce558SJens Axboe */ 934b92ce558SJens Axboe ret = 0; 935b92ce558SJens Axboe bytes = 0; 936c66ab6faSJens Axboe len = sd->total_len; 937c66ab6faSJens Axboe flags = sd->flags; 938c66ab6faSJens Axboe 939c66ab6faSJens Axboe /* 940c66ab6faSJens Axboe * Don't block on output, we have to drain the direct pipe. 941c66ab6faSJens Axboe */ 942c66ab6faSJens Axboe sd->flags &= ~SPLICE_F_NONBLOCK; 9430ff28d9fSChristophe Leroy more = sd->flags & SPLICE_F_MORE; 944b92ce558SJens Axboe 94517614445SDarrick J. Wong WARN_ON_ONCE(pipe->nrbufs != 0); 94617614445SDarrick J. Wong 947b92ce558SJens Axboe while (len) { 9483253d9d0SDarrick J. Wong unsigned int pipe_pages; 94951a92c0fSJens Axboe size_t read_len; 950a82c53a0STom Zanussi loff_t pos = sd->pos, prev_pos = pos; 951b92ce558SJens Axboe 95217614445SDarrick J. Wong /* Don't try to read more the pipe has space for. */ 9533253d9d0SDarrick J. Wong pipe_pages = pipe->buffers - pipe->nrbufs; 9543253d9d0SDarrick J. Wong read_len = min(len, (size_t)pipe_pages << PAGE_SHIFT); 95517614445SDarrick J. Wong ret = do_splice_to(in, &pos, pipe, read_len, flags); 95651a92c0fSJens Axboe if (unlikely(ret <= 0)) 957b92ce558SJens Axboe goto out_release; 958b92ce558SJens Axboe 959b92ce558SJens Axboe read_len = ret; 960c66ab6faSJens Axboe sd->total_len = read_len; 961b92ce558SJens Axboe 962b92ce558SJens Axboe /* 9630ff28d9fSChristophe Leroy * If more data is pending, set SPLICE_F_MORE 9640ff28d9fSChristophe Leroy * If this is the last data and SPLICE_F_MORE was not set 9650ff28d9fSChristophe Leroy * initially, clears it. 9660ff28d9fSChristophe Leroy */ 9670ff28d9fSChristophe Leroy if (read_len < len) 9680ff28d9fSChristophe Leroy sd->flags |= SPLICE_F_MORE; 9690ff28d9fSChristophe Leroy else if (!more) 9700ff28d9fSChristophe Leroy sd->flags &= ~SPLICE_F_MORE; 9710ff28d9fSChristophe Leroy /* 972b92ce558SJens Axboe * NOTE: nonblocking mode only applies to the input. We 973b92ce558SJens Axboe * must not do the output in nonblocking mode as then we 974b92ce558SJens Axboe * could get stuck data in the internal pipe: 975b92ce558SJens Axboe */ 976c66ab6faSJens Axboe ret = actor(pipe, sd); 977a82c53a0STom Zanussi if (unlikely(ret <= 0)) { 978a82c53a0STom Zanussi sd->pos = prev_pos; 979b92ce558SJens Axboe goto out_release; 980a82c53a0STom Zanussi } 981b92ce558SJens Axboe 982b92ce558SJens Axboe bytes += ret; 983b92ce558SJens Axboe len -= ret; 984bcd4f3acSJens Axboe sd->pos = pos; 985b92ce558SJens Axboe 986a82c53a0STom Zanussi if (ret < read_len) { 987a82c53a0STom Zanussi sd->pos = prev_pos + ret; 98851a92c0fSJens Axboe goto out_release; 989b92ce558SJens Axboe } 990a82c53a0STom Zanussi } 991b92ce558SJens Axboe 9929e97198dSJens Axboe done: 993b92ce558SJens Axboe pipe->nrbufs = pipe->curbuf = 0; 9949e97198dSJens Axboe file_accessed(in); 995b92ce558SJens Axboe return bytes; 996b92ce558SJens Axboe 997b92ce558SJens Axboe out_release: 998b92ce558SJens Axboe /* 999b92ce558SJens Axboe * If we did an incomplete transfer we must release 1000b92ce558SJens Axboe * the pipe buffers in question: 1001b92ce558SJens Axboe */ 100235f3d14dSJens Axboe for (i = 0; i < pipe->buffers; i++) { 1003b92ce558SJens Axboe struct pipe_buffer *buf = pipe->bufs + i; 1004b92ce558SJens Axboe 1005a779638cSMiklos Szeredi if (buf->ops) 1006a779638cSMiklos Szeredi pipe_buf_release(pipe, buf); 1007b92ce558SJens Axboe } 1008b92ce558SJens Axboe 10099e97198dSJens Axboe if (!bytes) 10109e97198dSJens Axboe bytes = ret; 1011b92ce558SJens Axboe 10129e97198dSJens Axboe goto done; 1013c66ab6faSJens Axboe } 1014c66ab6faSJens Axboe EXPORT_SYMBOL(splice_direct_to_actor); 1015c66ab6faSJens Axboe 1016c66ab6faSJens Axboe static int direct_splice_actor(struct pipe_inode_info *pipe, 1017c66ab6faSJens Axboe struct splice_desc *sd) 1018c66ab6faSJens Axboe { 10196a14b90bSJens Axboe struct file *file = sd->u.file; 1020c66ab6faSJens Axboe 10217995bd28SAl Viro return do_splice_from(pipe, file, sd->opos, sd->total_len, 10222cb4b05eSChangli Gao sd->flags); 1023c66ab6faSJens Axboe } 1024c66ab6faSJens Axboe 1025932cc6d4SJens Axboe /** 1026932cc6d4SJens Axboe * do_splice_direct - splices data directly between two files 1027932cc6d4SJens Axboe * @in: file to splice from 1028932cc6d4SJens Axboe * @ppos: input file offset 1029932cc6d4SJens Axboe * @out: file to splice to 1030acdb37c3SRandy Dunlap * @opos: output file offset 1031932cc6d4SJens Axboe * @len: number of bytes to splice 1032932cc6d4SJens Axboe * @flags: splice modifier flags 1033932cc6d4SJens Axboe * 1034932cc6d4SJens Axboe * Description: 1035932cc6d4SJens Axboe * For use by do_sendfile(). splice can easily emulate sendfile, but 1036932cc6d4SJens Axboe * doing it in the application would incur an extra system call 1037932cc6d4SJens Axboe * (splice in + splice out, as compared to just sendfile()). So this helper 1038932cc6d4SJens Axboe * can splice directly through a process-private pipe. 1039932cc6d4SJens Axboe * 1040932cc6d4SJens Axboe */ 1041c66ab6faSJens Axboe long do_splice_direct(struct file *in, loff_t *ppos, struct file *out, 10427995bd28SAl Viro loff_t *opos, size_t len, unsigned int flags) 1043c66ab6faSJens Axboe { 1044c66ab6faSJens Axboe struct splice_desc sd = { 1045c66ab6faSJens Axboe .len = len, 1046c66ab6faSJens Axboe .total_len = len, 1047c66ab6faSJens Axboe .flags = flags, 1048c66ab6faSJens Axboe .pos = *ppos, 10496a14b90bSJens Axboe .u.file = out, 10507995bd28SAl Viro .opos = opos, 1051c66ab6faSJens Axboe }; 105251a92c0fSJens Axboe long ret; 1053c66ab6faSJens Axboe 105418c67cb9SAl Viro if (unlikely(!(out->f_mode & FMODE_WRITE))) 105518c67cb9SAl Viro return -EBADF; 105618c67cb9SAl Viro 105718c67cb9SAl Viro if (unlikely(out->f_flags & O_APPEND)) 105818c67cb9SAl Viro return -EINVAL; 105918c67cb9SAl Viro 106018c67cb9SAl Viro ret = rw_verify_area(WRITE, out, opos, len); 106118c67cb9SAl Viro if (unlikely(ret < 0)) 106218c67cb9SAl Viro return ret; 106318c67cb9SAl Viro 1064c66ab6faSJens Axboe ret = splice_direct_to_actor(in, &sd, direct_splice_actor); 106551a92c0fSJens Axboe if (ret > 0) 1066a82c53a0STom Zanussi *ppos = sd.pos; 106751a92c0fSJens Axboe 1068c66ab6faSJens Axboe return ret; 1069b92ce558SJens Axboe } 10701c118596SMiklos Szeredi EXPORT_SYMBOL(do_splice_direct); 1071b92ce558SJens Axboe 10728924feffSAl Viro static int wait_for_space(struct pipe_inode_info *pipe, unsigned flags) 10738924feffSAl Viro { 107452bce911SLinus Torvalds for (;;) { 107552bce911SLinus Torvalds if (unlikely(!pipe->readers)) { 107652bce911SLinus Torvalds send_sig(SIGPIPE, current, 0); 107752bce911SLinus Torvalds return -EPIPE; 107852bce911SLinus Torvalds } 107952bce911SLinus Torvalds if (pipe->nrbufs != pipe->buffers) 108052bce911SLinus Torvalds return 0; 10818924feffSAl Viro if (flags & SPLICE_F_NONBLOCK) 10828924feffSAl Viro return -EAGAIN; 10838924feffSAl Viro if (signal_pending(current)) 10848924feffSAl Viro return -ERESTARTSYS; 10858924feffSAl Viro pipe->waiting_writers++; 10868924feffSAl Viro pipe_wait(pipe); 10878924feffSAl Viro pipe->waiting_writers--; 10888924feffSAl Viro } 10898924feffSAl Viro } 10908924feffSAl Viro 10917c77f0b3SMiklos Szeredi static int splice_pipe_to_pipe(struct pipe_inode_info *ipipe, 10927c77f0b3SMiklos Szeredi struct pipe_inode_info *opipe, 10937c77f0b3SMiklos Szeredi size_t len, unsigned int flags); 1094ddac0d39SJens Axboe 1095ddac0d39SJens Axboe /* 109683f9135bSJens Axboe * Determine where to splice to/from. 109783f9135bSJens Axboe */ 1098529565dcSIngo Molnar static long do_splice(struct file *in, loff_t __user *off_in, 1099529565dcSIngo Molnar struct file *out, loff_t __user *off_out, 1100529565dcSIngo Molnar size_t len, unsigned int flags) 11015274f052SJens Axboe { 11027c77f0b3SMiklos Szeredi struct pipe_inode_info *ipipe; 11037c77f0b3SMiklos Szeredi struct pipe_inode_info *opipe; 11047995bd28SAl Viro loff_t offset; 1105a4514ebdSJens Axboe long ret; 11065274f052SJens Axboe 110771993e62SLinus Torvalds ipipe = get_pipe_info(in); 110871993e62SLinus Torvalds opipe = get_pipe_info(out); 11097c77f0b3SMiklos Szeredi 11107c77f0b3SMiklos Szeredi if (ipipe && opipe) { 11117c77f0b3SMiklos Szeredi if (off_in || off_out) 11127c77f0b3SMiklos Szeredi return -ESPIPE; 11137c77f0b3SMiklos Szeredi 11147c77f0b3SMiklos Szeredi if (!(in->f_mode & FMODE_READ)) 11157c77f0b3SMiklos Szeredi return -EBADF; 11167c77f0b3SMiklos Szeredi 11177c77f0b3SMiklos Szeredi if (!(out->f_mode & FMODE_WRITE)) 11187c77f0b3SMiklos Szeredi return -EBADF; 11197c77f0b3SMiklos Szeredi 11207c77f0b3SMiklos Szeredi /* Splicing to self would be fun, but... */ 11217c77f0b3SMiklos Szeredi if (ipipe == opipe) 11227c77f0b3SMiklos Szeredi return -EINVAL; 11237c77f0b3SMiklos Szeredi 1124ee5e0011SSlavomir Kaslev if ((in->f_flags | out->f_flags) & O_NONBLOCK) 1125ee5e0011SSlavomir Kaslev flags |= SPLICE_F_NONBLOCK; 1126ee5e0011SSlavomir Kaslev 11277c77f0b3SMiklos Szeredi return splice_pipe_to_pipe(ipipe, opipe, len, flags); 11287c77f0b3SMiklos Szeredi } 11297c77f0b3SMiklos Szeredi 11307c77f0b3SMiklos Szeredi if (ipipe) { 1131529565dcSIngo Molnar if (off_in) 1132529565dcSIngo Molnar return -ESPIPE; 1133b92ce558SJens Axboe if (off_out) { 113419c9a49bSChangli Gao if (!(out->f_mode & FMODE_PWRITE)) 1135b92ce558SJens Axboe return -EINVAL; 1136cbb7e577SJens Axboe if (copy_from_user(&offset, off_out, sizeof(loff_t))) 1137b92ce558SJens Axboe return -EFAULT; 11387995bd28SAl Viro } else { 11397995bd28SAl Viro offset = out->f_pos; 11407995bd28SAl Viro } 1141529565dcSIngo Molnar 114218c67cb9SAl Viro if (unlikely(!(out->f_mode & FMODE_WRITE))) 114318c67cb9SAl Viro return -EBADF; 114418c67cb9SAl Viro 114518c67cb9SAl Viro if (unlikely(out->f_flags & O_APPEND)) 114618c67cb9SAl Viro return -EINVAL; 114718c67cb9SAl Viro 114818c67cb9SAl Viro ret = rw_verify_area(WRITE, out, &offset, len); 114918c67cb9SAl Viro if (unlikely(ret < 0)) 115018c67cb9SAl Viro return ret; 115118c67cb9SAl Viro 1152ee5e0011SSlavomir Kaslev if (in->f_flags & O_NONBLOCK) 1153ee5e0011SSlavomir Kaslev flags |= SPLICE_F_NONBLOCK; 1154ee5e0011SSlavomir Kaslev 1155500368f7SAl Viro file_start_write(out); 11567995bd28SAl Viro ret = do_splice_from(ipipe, out, &offset, len, flags); 1157500368f7SAl Viro file_end_write(out); 1158a4514ebdSJens Axboe 11597995bd28SAl Viro if (!off_out) 11607995bd28SAl Viro out->f_pos = offset; 11617995bd28SAl Viro else if (copy_to_user(off_out, &offset, sizeof(loff_t))) 1162a4514ebdSJens Axboe ret = -EFAULT; 1163a4514ebdSJens Axboe 1164a4514ebdSJens Axboe return ret; 1165529565dcSIngo Molnar } 11665274f052SJens Axboe 11677c77f0b3SMiklos Szeredi if (opipe) { 1168529565dcSIngo Molnar if (off_out) 1169529565dcSIngo Molnar return -ESPIPE; 1170b92ce558SJens Axboe if (off_in) { 117119c9a49bSChangli Gao if (!(in->f_mode & FMODE_PREAD)) 1172b92ce558SJens Axboe return -EINVAL; 1173cbb7e577SJens Axboe if (copy_from_user(&offset, off_in, sizeof(loff_t))) 1174b92ce558SJens Axboe return -EFAULT; 11757995bd28SAl Viro } else { 11767995bd28SAl Viro offset = in->f_pos; 11777995bd28SAl Viro } 1178529565dcSIngo Molnar 1179ee5e0011SSlavomir Kaslev if (out->f_flags & O_NONBLOCK) 1180ee5e0011SSlavomir Kaslev flags |= SPLICE_F_NONBLOCK; 1181ee5e0011SSlavomir Kaslev 11828924feffSAl Viro pipe_lock(opipe); 11838924feffSAl Viro ret = wait_for_space(opipe, flags); 11843253d9d0SDarrick J. Wong if (!ret) { 11853253d9d0SDarrick J. Wong unsigned int pipe_pages; 11863253d9d0SDarrick J. Wong 11873253d9d0SDarrick J. Wong /* Don't try to read more the pipe has space for. */ 11883253d9d0SDarrick J. Wong pipe_pages = opipe->buffers - opipe->nrbufs; 11893253d9d0SDarrick J. Wong len = min(len, (size_t)pipe_pages << PAGE_SHIFT); 11903253d9d0SDarrick J. Wong 11917995bd28SAl Viro ret = do_splice_to(in, &offset, opipe, len, flags); 11923253d9d0SDarrick J. Wong } 11938924feffSAl Viro pipe_unlock(opipe); 11948924feffSAl Viro if (ret > 0) 11958924feffSAl Viro wakeup_pipe_readers(opipe); 11967995bd28SAl Viro if (!off_in) 11977995bd28SAl Viro in->f_pos = offset; 11987995bd28SAl Viro else if (copy_to_user(off_in, &offset, sizeof(loff_t))) 1199a4514ebdSJens Axboe ret = -EFAULT; 1200a4514ebdSJens Axboe 1201a4514ebdSJens Axboe return ret; 1202529565dcSIngo Molnar } 12035274f052SJens Axboe 12045274f052SJens Axboe return -EINVAL; 12055274f052SJens Axboe } 12065274f052SJens Axboe 120779fddc4eSAl Viro static int iter_to_pipe(struct iov_iter *from, 120879fddc4eSAl Viro struct pipe_inode_info *pipe, 120979fddc4eSAl Viro unsigned flags) 1210912d35f8SJens Axboe { 121179fddc4eSAl Viro struct pipe_buffer buf = { 121279fddc4eSAl Viro .ops = &user_page_pipe_buf_ops, 121379fddc4eSAl Viro .flags = flags 121479fddc4eSAl Viro }; 121579fddc4eSAl Viro size_t total = 0; 121679fddc4eSAl Viro int ret = 0; 121779fddc4eSAl Viro bool failed = false; 121879fddc4eSAl Viro 121979fddc4eSAl Viro while (iov_iter_count(from) && !failed) { 122079fddc4eSAl Viro struct page *pages[16]; 1221db85a9ebSAl Viro ssize_t copied; 1222db85a9ebSAl Viro size_t start; 122379fddc4eSAl Viro int n; 1224912d35f8SJens Axboe 122579fddc4eSAl Viro copied = iov_iter_get_pages(from, pages, ~0UL, 16, &start); 122679fddc4eSAl Viro if (copied <= 0) { 122779fddc4eSAl Viro ret = copied; 122879fddc4eSAl Viro break; 122979fddc4eSAl Viro } 1230912d35f8SJens Axboe 123179fddc4eSAl Viro for (n = 0; copied; n++, start = 0) { 1232db85a9ebSAl Viro int size = min_t(int, copied, PAGE_SIZE - start); 123379fddc4eSAl Viro if (!failed) { 123479fddc4eSAl Viro buf.page = pages[n]; 123579fddc4eSAl Viro buf.offset = start; 123679fddc4eSAl Viro buf.len = size; 123779fddc4eSAl Viro ret = add_to_pipe(pipe, &buf); 123879fddc4eSAl Viro if (unlikely(ret < 0)) { 123979fddc4eSAl Viro failed = true; 124079fddc4eSAl Viro } else { 124179fddc4eSAl Viro iov_iter_advance(from, ret); 124279fddc4eSAl Viro total += ret; 124379fddc4eSAl Viro } 124479fddc4eSAl Viro } else { 124579fddc4eSAl Viro put_page(pages[n]); 124679fddc4eSAl Viro } 1247db85a9ebSAl Viro copied -= size; 1248912d35f8SJens Axboe } 1249912d35f8SJens Axboe } 125079fddc4eSAl Viro return total ? total : ret; 1251912d35f8SJens Axboe } 1252912d35f8SJens Axboe 12536a14b90bSJens Axboe static int pipe_to_user(struct pipe_inode_info *pipe, struct pipe_buffer *buf, 12546a14b90bSJens Axboe struct splice_desc *sd) 12556a14b90bSJens Axboe { 12566130f531SAl Viro int n = copy_page_to_iter(buf->page, buf->offset, sd->len, sd->u.data); 12576130f531SAl Viro return n == sd->len ? n : -EFAULT; 12586a14b90bSJens Axboe } 12596a14b90bSJens Axboe 12606a14b90bSJens Axboe /* 12616a14b90bSJens Axboe * For lack of a better implementation, implement vmsplice() to userspace 12626a14b90bSJens Axboe * as a simple copy of the pipes pages to the user iov. 12636a14b90bSJens Axboe */ 126487a3002aSAl Viro static long vmsplice_to_user(struct file *file, struct iov_iter *iter, 126587a3002aSAl Viro unsigned int flags) 12666a14b90bSJens Axboe { 126787a3002aSAl Viro struct pipe_inode_info *pipe = get_pipe_info(file); 126887a3002aSAl Viro struct splice_desc sd = { 126987a3002aSAl Viro .total_len = iov_iter_count(iter), 127087a3002aSAl Viro .flags = flags, 127187a3002aSAl Viro .u.data = iter 127287a3002aSAl Viro }; 127387a3002aSAl Viro long ret = 0; 12746a14b90bSJens Axboe 12756a14b90bSJens Axboe if (!pipe) 12766a14b90bSJens Axboe return -EBADF; 12776a14b90bSJens Axboe 1278345995faSAl Viro if (sd.total_len) { 12796130f531SAl Viro pipe_lock(pipe); 12806130f531SAl Viro ret = __splice_from_pipe(pipe, &sd, pipe_to_user); 128161e0d47cSMiklos Szeredi pipe_unlock(pipe); 1282345995faSAl Viro } 12836a14b90bSJens Axboe 12846a14b90bSJens Axboe return ret; 12856a14b90bSJens Axboe } 12866a14b90bSJens Axboe 1287912d35f8SJens Axboe /* 1288912d35f8SJens Axboe * vmsplice splices a user address range into a pipe. It can be thought of 1289912d35f8SJens Axboe * as splice-from-memory, where the regular splice is splice-from-file (or 1290912d35f8SJens Axboe * to file). In both cases the output is a pipe, naturally. 1291912d35f8SJens Axboe */ 129287a3002aSAl Viro static long vmsplice_to_pipe(struct file *file, struct iov_iter *iter, 129387a3002aSAl Viro unsigned int flags) 1294912d35f8SJens Axboe { 1295ddac0d39SJens Axboe struct pipe_inode_info *pipe; 129687a3002aSAl Viro long ret = 0; 129779fddc4eSAl Viro unsigned buf_flag = 0; 129879fddc4eSAl Viro 129979fddc4eSAl Viro if (flags & SPLICE_F_GIFT) 130079fddc4eSAl Viro buf_flag = PIPE_BUF_FLAG_GIFT; 1301912d35f8SJens Axboe 130271993e62SLinus Torvalds pipe = get_pipe_info(file); 1303ddac0d39SJens Axboe if (!pipe) 1304912d35f8SJens Axboe return -EBADF; 1305912d35f8SJens Axboe 13068924feffSAl Viro pipe_lock(pipe); 13078924feffSAl Viro ret = wait_for_space(pipe, flags); 130879fddc4eSAl Viro if (!ret) 130987a3002aSAl Viro ret = iter_to_pipe(iter, pipe, buf_flag); 13108924feffSAl Viro pipe_unlock(pipe); 13118924feffSAl Viro if (ret > 0) 13128924feffSAl Viro wakeup_pipe_readers(pipe); 131335f3d14dSJens Axboe return ret; 1314912d35f8SJens Axboe } 1315912d35f8SJens Axboe 131687a3002aSAl Viro static int vmsplice_type(struct fd f, int *type) 131787a3002aSAl Viro { 131887a3002aSAl Viro if (!f.file) 131987a3002aSAl Viro return -EBADF; 132087a3002aSAl Viro if (f.file->f_mode & FMODE_WRITE) { 132187a3002aSAl Viro *type = WRITE; 132287a3002aSAl Viro } else if (f.file->f_mode & FMODE_READ) { 132387a3002aSAl Viro *type = READ; 132487a3002aSAl Viro } else { 132587a3002aSAl Viro fdput(f); 132687a3002aSAl Viro return -EBADF; 132787a3002aSAl Viro } 132887a3002aSAl Viro return 0; 132987a3002aSAl Viro } 133087a3002aSAl Viro 13316a14b90bSJens Axboe /* 13326a14b90bSJens Axboe * Note that vmsplice only really supports true splicing _from_ user memory 13336a14b90bSJens Axboe * to a pipe, not the other way around. Splicing from user memory is a simple 13346a14b90bSJens Axboe * operation that can be supported without any funky alignment restrictions 13356a14b90bSJens Axboe * or nasty vm tricks. We simply map in the user memory and fill them into 13366a14b90bSJens Axboe * a pipe. The reverse isn't quite as easy, though. There are two possible 13376a14b90bSJens Axboe * solutions for that: 13386a14b90bSJens Axboe * 13396a14b90bSJens Axboe * - memcpy() the data internally, at which point we might as well just 13406a14b90bSJens Axboe * do a regular read() on the buffer anyway. 13416a14b90bSJens Axboe * - Lots of nasty vm tricks, that are neither fast nor flexible (it 13426a14b90bSJens Axboe * has restriction limitations on both ends of the pipe). 13436a14b90bSJens Axboe * 13446a14b90bSJens Axboe * Currently we punt and implement it as a normal copy, see pipe_to_user(). 13456a14b90bSJens Axboe * 13466a14b90bSJens Axboe */ 134787a3002aSAl Viro static long do_vmsplice(struct file *f, struct iov_iter *iter, unsigned int flags) 1348912d35f8SJens Axboe { 13493d6ea290SAl Viro if (unlikely(flags & ~SPLICE_F_ALL)) 13503d6ea290SAl Viro return -EINVAL; 135187a3002aSAl Viro 135287a3002aSAl Viro if (!iov_iter_count(iter)) 13536a14b90bSJens Axboe return 0; 13546a14b90bSJens Axboe 135587a3002aSAl Viro if (iov_iter_rw(iter) == WRITE) 135687a3002aSAl Viro return vmsplice_to_pipe(f, iter, flags); 135787a3002aSAl Viro else 135887a3002aSAl Viro return vmsplice_to_user(f, iter, flags); 1359912d35f8SJens Axboe } 1360912d35f8SJens Axboe 136187a3002aSAl Viro SYSCALL_DEFINE4(vmsplice, int, fd, const struct iovec __user *, uiov, 136230cfe4efSDominik Brodowski unsigned long, nr_segs, unsigned int, flags) 136330cfe4efSDominik Brodowski { 136487a3002aSAl Viro struct iovec iovstack[UIO_FASTIOV]; 136587a3002aSAl Viro struct iovec *iov = iovstack; 136687a3002aSAl Viro struct iov_iter iter; 136787e5e6daSJens Axboe ssize_t error; 136887a3002aSAl Viro struct fd f; 136987a3002aSAl Viro int type; 137087a3002aSAl Viro 137187a3002aSAl Viro f = fdget(fd); 137287a3002aSAl Viro error = vmsplice_type(f, &type); 137387a3002aSAl Viro if (error) 137487a3002aSAl Viro return error; 137587a3002aSAl Viro 137687a3002aSAl Viro error = import_iovec(type, uiov, nr_segs, 137787a3002aSAl Viro ARRAY_SIZE(iovstack), &iov, &iter); 137887e5e6daSJens Axboe if (error >= 0) { 137987a3002aSAl Viro error = do_vmsplice(f.file, &iter, flags); 138087a3002aSAl Viro kfree(iov); 138187a3002aSAl Viro } 138287a3002aSAl Viro fdput(f); 138387a3002aSAl Viro return error; 138430cfe4efSDominik Brodowski } 138530cfe4efSDominik Brodowski 138676b021d0SAl Viro #ifdef CONFIG_COMPAT 138776b021d0SAl Viro COMPAT_SYSCALL_DEFINE4(vmsplice, int, fd, const struct compat_iovec __user *, iov32, 138876b021d0SAl Viro unsigned int, nr_segs, unsigned int, flags) 138976b021d0SAl Viro { 139087a3002aSAl Viro struct iovec iovstack[UIO_FASTIOV]; 139187a3002aSAl Viro struct iovec *iov = iovstack; 139287a3002aSAl Viro struct iov_iter iter; 139387e5e6daSJens Axboe ssize_t error; 139487a3002aSAl Viro struct fd f; 139587a3002aSAl Viro int type; 139687a3002aSAl Viro 139787a3002aSAl Viro f = fdget(fd); 139887a3002aSAl Viro error = vmsplice_type(f, &type); 139987a3002aSAl Viro if (error) 140087a3002aSAl Viro return error; 140187a3002aSAl Viro 140287a3002aSAl Viro error = compat_import_iovec(type, iov32, nr_segs, 140387a3002aSAl Viro ARRAY_SIZE(iovstack), &iov, &iter); 140487e5e6daSJens Axboe if (error >= 0) { 140587a3002aSAl Viro error = do_vmsplice(f.file, &iter, flags); 140687a3002aSAl Viro kfree(iov); 140776b021d0SAl Viro } 140887a3002aSAl Viro fdput(f); 140987a3002aSAl Viro return error; 141076b021d0SAl Viro } 141176b021d0SAl Viro #endif 141276b021d0SAl Viro 1413836f92adSHeiko Carstens SYSCALL_DEFINE6(splice, int, fd_in, loff_t __user *, off_in, 1414836f92adSHeiko Carstens int, fd_out, loff_t __user *, off_out, 1415836f92adSHeiko Carstens size_t, len, unsigned int, flags) 14165274f052SJens Axboe { 14172903ff01SAl Viro struct fd in, out; 14185274f052SJens Axboe long error; 14195274f052SJens Axboe 14205274f052SJens Axboe if (unlikely(!len)) 14215274f052SJens Axboe return 0; 14225274f052SJens Axboe 14233d6ea290SAl Viro if (unlikely(flags & ~SPLICE_F_ALL)) 14243d6ea290SAl Viro return -EINVAL; 14253d6ea290SAl Viro 14265274f052SJens Axboe error = -EBADF; 14272903ff01SAl Viro in = fdget(fd_in); 14282903ff01SAl Viro if (in.file) { 14292903ff01SAl Viro if (in.file->f_mode & FMODE_READ) { 14302903ff01SAl Viro out = fdget(fd_out); 14312903ff01SAl Viro if (out.file) { 14322903ff01SAl Viro if (out.file->f_mode & FMODE_WRITE) 14332903ff01SAl Viro error = do_splice(in.file, off_in, 14342903ff01SAl Viro out.file, off_out, 1435529565dcSIngo Molnar len, flags); 14362903ff01SAl Viro fdput(out); 14375274f052SJens Axboe } 14385274f052SJens Axboe } 14392903ff01SAl Viro fdput(in); 14405274f052SJens Axboe } 14415274f052SJens Axboe return error; 14425274f052SJens Axboe } 144370524490SJens Axboe 144470524490SJens Axboe /* 1445aadd06e5SJens Axboe * Make sure there's data to read. Wait for input if we can, otherwise 1446aadd06e5SJens Axboe * return an appropriate error. 1447aadd06e5SJens Axboe */ 14487c77f0b3SMiklos Szeredi static int ipipe_prep(struct pipe_inode_info *pipe, unsigned int flags) 1449aadd06e5SJens Axboe { 1450aadd06e5SJens Axboe int ret; 1451aadd06e5SJens Axboe 1452aadd06e5SJens Axboe /* 1453aadd06e5SJens Axboe * Check ->nrbufs without the inode lock first. This function 1454aadd06e5SJens Axboe * is speculative anyways, so missing one is ok. 1455aadd06e5SJens Axboe */ 1456aadd06e5SJens Axboe if (pipe->nrbufs) 1457aadd06e5SJens Axboe return 0; 1458aadd06e5SJens Axboe 1459aadd06e5SJens Axboe ret = 0; 146061e0d47cSMiklos Szeredi pipe_lock(pipe); 1461aadd06e5SJens Axboe 1462aadd06e5SJens Axboe while (!pipe->nrbufs) { 1463aadd06e5SJens Axboe if (signal_pending(current)) { 1464aadd06e5SJens Axboe ret = -ERESTARTSYS; 1465aadd06e5SJens Axboe break; 1466aadd06e5SJens Axboe } 1467aadd06e5SJens Axboe if (!pipe->writers) 1468aadd06e5SJens Axboe break; 1469aadd06e5SJens Axboe if (!pipe->waiting_writers) { 1470aadd06e5SJens Axboe if (flags & SPLICE_F_NONBLOCK) { 1471aadd06e5SJens Axboe ret = -EAGAIN; 1472aadd06e5SJens Axboe break; 1473aadd06e5SJens Axboe } 1474aadd06e5SJens Axboe } 1475aadd06e5SJens Axboe pipe_wait(pipe); 1476aadd06e5SJens Axboe } 1477aadd06e5SJens Axboe 147861e0d47cSMiklos Szeredi pipe_unlock(pipe); 1479aadd06e5SJens Axboe return ret; 1480aadd06e5SJens Axboe } 1481aadd06e5SJens Axboe 1482aadd06e5SJens Axboe /* 1483aadd06e5SJens Axboe * Make sure there's writeable room. Wait for room if we can, otherwise 1484aadd06e5SJens Axboe * return an appropriate error. 1485aadd06e5SJens Axboe */ 14867c77f0b3SMiklos Szeredi static int opipe_prep(struct pipe_inode_info *pipe, unsigned int flags) 1487aadd06e5SJens Axboe { 1488aadd06e5SJens Axboe int ret; 1489aadd06e5SJens Axboe 1490aadd06e5SJens Axboe /* 1491aadd06e5SJens Axboe * Check ->nrbufs without the inode lock first. This function 1492aadd06e5SJens Axboe * is speculative anyways, so missing one is ok. 1493aadd06e5SJens Axboe */ 149435f3d14dSJens Axboe if (pipe->nrbufs < pipe->buffers) 1495aadd06e5SJens Axboe return 0; 1496aadd06e5SJens Axboe 1497aadd06e5SJens Axboe ret = 0; 149861e0d47cSMiklos Szeredi pipe_lock(pipe); 1499aadd06e5SJens Axboe 150035f3d14dSJens Axboe while (pipe->nrbufs >= pipe->buffers) { 1501aadd06e5SJens Axboe if (!pipe->readers) { 1502aadd06e5SJens Axboe send_sig(SIGPIPE, current, 0); 1503aadd06e5SJens Axboe ret = -EPIPE; 1504aadd06e5SJens Axboe break; 1505aadd06e5SJens Axboe } 1506aadd06e5SJens Axboe if (flags & SPLICE_F_NONBLOCK) { 1507aadd06e5SJens Axboe ret = -EAGAIN; 1508aadd06e5SJens Axboe break; 1509aadd06e5SJens Axboe } 1510aadd06e5SJens Axboe if (signal_pending(current)) { 1511aadd06e5SJens Axboe ret = -ERESTARTSYS; 1512aadd06e5SJens Axboe break; 1513aadd06e5SJens Axboe } 1514aadd06e5SJens Axboe pipe->waiting_writers++; 1515aadd06e5SJens Axboe pipe_wait(pipe); 1516aadd06e5SJens Axboe pipe->waiting_writers--; 1517aadd06e5SJens Axboe } 1518aadd06e5SJens Axboe 151961e0d47cSMiklos Szeredi pipe_unlock(pipe); 1520aadd06e5SJens Axboe return ret; 1521aadd06e5SJens Axboe } 1522aadd06e5SJens Axboe 1523aadd06e5SJens Axboe /* 15247c77f0b3SMiklos Szeredi * Splice contents of ipipe to opipe. 15257c77f0b3SMiklos Szeredi */ 15267c77f0b3SMiklos Szeredi static int splice_pipe_to_pipe(struct pipe_inode_info *ipipe, 15277c77f0b3SMiklos Szeredi struct pipe_inode_info *opipe, 15287c77f0b3SMiklos Szeredi size_t len, unsigned int flags) 15297c77f0b3SMiklos Szeredi { 15307c77f0b3SMiklos Szeredi struct pipe_buffer *ibuf, *obuf; 15317c77f0b3SMiklos Szeredi int ret = 0, nbuf; 15327c77f0b3SMiklos Szeredi bool input_wakeup = false; 15337c77f0b3SMiklos Szeredi 15347c77f0b3SMiklos Szeredi 15357c77f0b3SMiklos Szeredi retry: 15367c77f0b3SMiklos Szeredi ret = ipipe_prep(ipipe, flags); 15377c77f0b3SMiklos Szeredi if (ret) 15387c77f0b3SMiklos Szeredi return ret; 15397c77f0b3SMiklos Szeredi 15407c77f0b3SMiklos Szeredi ret = opipe_prep(opipe, flags); 15417c77f0b3SMiklos Szeredi if (ret) 15427c77f0b3SMiklos Szeredi return ret; 15437c77f0b3SMiklos Szeredi 15447c77f0b3SMiklos Szeredi /* 15457c77f0b3SMiklos Szeredi * Potential ABBA deadlock, work around it by ordering lock 15467c77f0b3SMiklos Szeredi * grabbing by pipe info address. Otherwise two different processes 15477c77f0b3SMiklos Szeredi * could deadlock (one doing tee from A -> B, the other from B -> A). 15487c77f0b3SMiklos Szeredi */ 15497c77f0b3SMiklos Szeredi pipe_double_lock(ipipe, opipe); 15507c77f0b3SMiklos Szeredi 15517c77f0b3SMiklos Szeredi do { 15527c77f0b3SMiklos Szeredi if (!opipe->readers) { 15537c77f0b3SMiklos Szeredi send_sig(SIGPIPE, current, 0); 15547c77f0b3SMiklos Szeredi if (!ret) 15557c77f0b3SMiklos Szeredi ret = -EPIPE; 15567c77f0b3SMiklos Szeredi break; 15577c77f0b3SMiklos Szeredi } 15587c77f0b3SMiklos Szeredi 15597c77f0b3SMiklos Szeredi if (!ipipe->nrbufs && !ipipe->writers) 15607c77f0b3SMiklos Szeredi break; 15617c77f0b3SMiklos Szeredi 15627c77f0b3SMiklos Szeredi /* 15637c77f0b3SMiklos Szeredi * Cannot make any progress, because either the input 15647c77f0b3SMiklos Szeredi * pipe is empty or the output pipe is full. 15657c77f0b3SMiklos Szeredi */ 156635f3d14dSJens Axboe if (!ipipe->nrbufs || opipe->nrbufs >= opipe->buffers) { 15677c77f0b3SMiklos Szeredi /* Already processed some buffers, break */ 15687c77f0b3SMiklos Szeredi if (ret) 15697c77f0b3SMiklos Szeredi break; 15707c77f0b3SMiklos Szeredi 15717c77f0b3SMiklos Szeredi if (flags & SPLICE_F_NONBLOCK) { 15727c77f0b3SMiklos Szeredi ret = -EAGAIN; 15737c77f0b3SMiklos Szeredi break; 15747c77f0b3SMiklos Szeredi } 15757c77f0b3SMiklos Szeredi 15767c77f0b3SMiklos Szeredi /* 15777c77f0b3SMiklos Szeredi * We raced with another reader/writer and haven't 15787c77f0b3SMiklos Szeredi * managed to process any buffers. A zero return 15797c77f0b3SMiklos Szeredi * value means EOF, so retry instead. 15807c77f0b3SMiklos Szeredi */ 15817c77f0b3SMiklos Szeredi pipe_unlock(ipipe); 15827c77f0b3SMiklos Szeredi pipe_unlock(opipe); 15837c77f0b3SMiklos Szeredi goto retry; 15847c77f0b3SMiklos Szeredi } 15857c77f0b3SMiklos Szeredi 15867c77f0b3SMiklos Szeredi ibuf = ipipe->bufs + ipipe->curbuf; 158735f3d14dSJens Axboe nbuf = (opipe->curbuf + opipe->nrbufs) & (opipe->buffers - 1); 15887c77f0b3SMiklos Szeredi obuf = opipe->bufs + nbuf; 15897c77f0b3SMiklos Szeredi 15907c77f0b3SMiklos Szeredi if (len >= ibuf->len) { 15917c77f0b3SMiklos Szeredi /* 15927c77f0b3SMiklos Szeredi * Simply move the whole buffer from ipipe to opipe 15937c77f0b3SMiklos Szeredi */ 15947c77f0b3SMiklos Szeredi *obuf = *ibuf; 15957c77f0b3SMiklos Szeredi ibuf->ops = NULL; 15967c77f0b3SMiklos Szeredi opipe->nrbufs++; 159735f3d14dSJens Axboe ipipe->curbuf = (ipipe->curbuf + 1) & (ipipe->buffers - 1); 15987c77f0b3SMiklos Szeredi ipipe->nrbufs--; 15997c77f0b3SMiklos Szeredi input_wakeup = true; 16007c77f0b3SMiklos Szeredi } else { 16017c77f0b3SMiklos Szeredi /* 16027c77f0b3SMiklos Szeredi * Get a reference to this pipe buffer, 16037c77f0b3SMiklos Szeredi * so we can copy the contents over. 16047c77f0b3SMiklos Szeredi */ 160515fab63eSMatthew Wilcox if (!pipe_buf_get(ipipe, ibuf)) { 160615fab63eSMatthew Wilcox if (ret == 0) 160715fab63eSMatthew Wilcox ret = -EFAULT; 160815fab63eSMatthew Wilcox break; 160915fab63eSMatthew Wilcox } 16107c77f0b3SMiklos Szeredi *obuf = *ibuf; 16117c77f0b3SMiklos Szeredi 16127c77f0b3SMiklos Szeredi /* 16137c77f0b3SMiklos Szeredi * Don't inherit the gift flag, we need to 16147c77f0b3SMiklos Szeredi * prevent multiple steals of this page. 16157c77f0b3SMiklos Szeredi */ 16167c77f0b3SMiklos Szeredi obuf->flags &= ~PIPE_BUF_FLAG_GIFT; 16177c77f0b3SMiklos Szeredi 1618a0ce2f0aSJann Horn pipe_buf_mark_unmergeable(obuf); 1619a0ce2f0aSJann Horn 16207c77f0b3SMiklos Szeredi obuf->len = len; 16217c77f0b3SMiklos Szeredi opipe->nrbufs++; 16227c77f0b3SMiklos Szeredi ibuf->offset += obuf->len; 16237c77f0b3SMiklos Szeredi ibuf->len -= obuf->len; 16247c77f0b3SMiklos Szeredi } 16257c77f0b3SMiklos Szeredi ret += obuf->len; 16267c77f0b3SMiklos Szeredi len -= obuf->len; 16277c77f0b3SMiklos Szeredi } while (len); 16287c77f0b3SMiklos Szeredi 16297c77f0b3SMiklos Szeredi pipe_unlock(ipipe); 16307c77f0b3SMiklos Szeredi pipe_unlock(opipe); 16317c77f0b3SMiklos Szeredi 16327c77f0b3SMiklos Szeredi /* 16337c77f0b3SMiklos Szeredi * If we put data in the output pipe, wakeup any potential readers. 16347c77f0b3SMiklos Szeredi */ 1635825cdcb1SNamhyung Kim if (ret > 0) 1636825cdcb1SNamhyung Kim wakeup_pipe_readers(opipe); 1637825cdcb1SNamhyung Kim 16387c77f0b3SMiklos Szeredi if (input_wakeup) 16397c77f0b3SMiklos Szeredi wakeup_pipe_writers(ipipe); 16407c77f0b3SMiklos Szeredi 16417c77f0b3SMiklos Szeredi return ret; 16427c77f0b3SMiklos Szeredi } 16437c77f0b3SMiklos Szeredi 16447c77f0b3SMiklos Szeredi /* 164570524490SJens Axboe * Link contents of ipipe to opipe. 164670524490SJens Axboe */ 164770524490SJens Axboe static int link_pipe(struct pipe_inode_info *ipipe, 164870524490SJens Axboe struct pipe_inode_info *opipe, 164970524490SJens Axboe size_t len, unsigned int flags) 165070524490SJens Axboe { 165170524490SJens Axboe struct pipe_buffer *ibuf, *obuf; 1652aadd06e5SJens Axboe int ret = 0, i = 0, nbuf; 165370524490SJens Axboe 165470524490SJens Axboe /* 165570524490SJens Axboe * Potential ABBA deadlock, work around it by ordering lock 165661e0d47cSMiklos Szeredi * grabbing by pipe info address. Otherwise two different processes 165770524490SJens Axboe * could deadlock (one doing tee from A -> B, the other from B -> A). 165870524490SJens Axboe */ 165961e0d47cSMiklos Szeredi pipe_double_lock(ipipe, opipe); 166070524490SJens Axboe 1661aadd06e5SJens Axboe do { 166270524490SJens Axboe if (!opipe->readers) { 166370524490SJens Axboe send_sig(SIGPIPE, current, 0); 166470524490SJens Axboe if (!ret) 166570524490SJens Axboe ret = -EPIPE; 166670524490SJens Axboe break; 166770524490SJens Axboe } 166870524490SJens Axboe 166970524490SJens Axboe /* 1670aadd06e5SJens Axboe * If we have iterated all input buffers or ran out of 1671aadd06e5SJens Axboe * output room, break. 167270524490SJens Axboe */ 167335f3d14dSJens Axboe if (i >= ipipe->nrbufs || opipe->nrbufs >= opipe->buffers) 1674aadd06e5SJens Axboe break; 1675aadd06e5SJens Axboe 167635f3d14dSJens Axboe ibuf = ipipe->bufs + ((ipipe->curbuf + i) & (ipipe->buffers-1)); 167735f3d14dSJens Axboe nbuf = (opipe->curbuf + opipe->nrbufs) & (opipe->buffers - 1); 167870524490SJens Axboe 167970524490SJens Axboe /* 168070524490SJens Axboe * Get a reference to this pipe buffer, 168170524490SJens Axboe * so we can copy the contents over. 168270524490SJens Axboe */ 168315fab63eSMatthew Wilcox if (!pipe_buf_get(ipipe, ibuf)) { 168415fab63eSMatthew Wilcox if (ret == 0) 168515fab63eSMatthew Wilcox ret = -EFAULT; 168615fab63eSMatthew Wilcox break; 168715fab63eSMatthew Wilcox } 168870524490SJens Axboe 168970524490SJens Axboe obuf = opipe->bufs + nbuf; 169070524490SJens Axboe *obuf = *ibuf; 169170524490SJens Axboe 16927afa6fd0SJens Axboe /* 16937afa6fd0SJens Axboe * Don't inherit the gift flag, we need to 16947afa6fd0SJens Axboe * prevent multiple steals of this page. 16957afa6fd0SJens Axboe */ 16967afa6fd0SJens Axboe obuf->flags &= ~PIPE_BUF_FLAG_GIFT; 16977afa6fd0SJens Axboe 1698a0ce2f0aSJann Horn pipe_buf_mark_unmergeable(obuf); 1699a0ce2f0aSJann Horn 170070524490SJens Axboe if (obuf->len > len) 170170524490SJens Axboe obuf->len = len; 170270524490SJens Axboe 170370524490SJens Axboe opipe->nrbufs++; 170470524490SJens Axboe ret += obuf->len; 170570524490SJens Axboe len -= obuf->len; 1706aadd06e5SJens Axboe i++; 1707aadd06e5SJens Axboe } while (len); 170870524490SJens Axboe 170902cf01aeSJens Axboe /* 171002cf01aeSJens Axboe * return EAGAIN if we have the potential of some data in the 171102cf01aeSJens Axboe * future, otherwise just return 0 171202cf01aeSJens Axboe */ 171302cf01aeSJens Axboe if (!ret && ipipe->waiting_writers && (flags & SPLICE_F_NONBLOCK)) 171402cf01aeSJens Axboe ret = -EAGAIN; 171502cf01aeSJens Axboe 171661e0d47cSMiklos Szeredi pipe_unlock(ipipe); 171761e0d47cSMiklos Szeredi pipe_unlock(opipe); 171870524490SJens Axboe 1719aadd06e5SJens Axboe /* 1720aadd06e5SJens Axboe * If we put data in the output pipe, wakeup any potential readers. 1721aadd06e5SJens Axboe */ 1722825cdcb1SNamhyung Kim if (ret > 0) 1723825cdcb1SNamhyung Kim wakeup_pipe_readers(opipe); 172470524490SJens Axboe 172570524490SJens Axboe return ret; 172670524490SJens Axboe } 172770524490SJens Axboe 172870524490SJens Axboe /* 172970524490SJens Axboe * This is a tee(1) implementation that works on pipes. It doesn't copy 173070524490SJens Axboe * any data, it simply references the 'in' pages on the 'out' pipe. 173170524490SJens Axboe * The 'flags' used are the SPLICE_F_* variants, currently the only 173270524490SJens Axboe * applicable one is SPLICE_F_NONBLOCK. 173370524490SJens Axboe */ 173470524490SJens Axboe static long do_tee(struct file *in, struct file *out, size_t len, 173570524490SJens Axboe unsigned int flags) 173670524490SJens Axboe { 173771993e62SLinus Torvalds struct pipe_inode_info *ipipe = get_pipe_info(in); 173871993e62SLinus Torvalds struct pipe_inode_info *opipe = get_pipe_info(out); 1739aadd06e5SJens Axboe int ret = -EINVAL; 174070524490SJens Axboe 174170524490SJens Axboe /* 1742aadd06e5SJens Axboe * Duplicate the contents of ipipe to opipe without actually 1743aadd06e5SJens Axboe * copying the data. 174470524490SJens Axboe */ 1745aadd06e5SJens Axboe if (ipipe && opipe && ipipe != opipe) { 1746ee5e0011SSlavomir Kaslev if ((in->f_flags | out->f_flags) & O_NONBLOCK) 1747ee5e0011SSlavomir Kaslev flags |= SPLICE_F_NONBLOCK; 1748ee5e0011SSlavomir Kaslev 1749aadd06e5SJens Axboe /* 1750aadd06e5SJens Axboe * Keep going, unless we encounter an error. The ipipe/opipe 1751aadd06e5SJens Axboe * ordering doesn't really matter. 1752aadd06e5SJens Axboe */ 17537c77f0b3SMiklos Szeredi ret = ipipe_prep(ipipe, flags); 1754aadd06e5SJens Axboe if (!ret) { 17557c77f0b3SMiklos Szeredi ret = opipe_prep(opipe, flags); 175602cf01aeSJens Axboe if (!ret) 1757aadd06e5SJens Axboe ret = link_pipe(ipipe, opipe, len, flags); 1758aadd06e5SJens Axboe } 1759aadd06e5SJens Axboe } 176070524490SJens Axboe 1761aadd06e5SJens Axboe return ret; 176270524490SJens Axboe } 176370524490SJens Axboe 1764836f92adSHeiko Carstens SYSCALL_DEFINE4(tee, int, fdin, int, fdout, size_t, len, unsigned int, flags) 176570524490SJens Axboe { 17662903ff01SAl Viro struct fd in; 17672903ff01SAl Viro int error; 176870524490SJens Axboe 17693d6ea290SAl Viro if (unlikely(flags & ~SPLICE_F_ALL)) 17703d6ea290SAl Viro return -EINVAL; 17713d6ea290SAl Viro 177270524490SJens Axboe if (unlikely(!len)) 177370524490SJens Axboe return 0; 177470524490SJens Axboe 177570524490SJens Axboe error = -EBADF; 17762903ff01SAl Viro in = fdget(fdin); 17772903ff01SAl Viro if (in.file) { 17782903ff01SAl Viro if (in.file->f_mode & FMODE_READ) { 17792903ff01SAl Viro struct fd out = fdget(fdout); 17802903ff01SAl Viro if (out.file) { 17812903ff01SAl Viro if (out.file->f_mode & FMODE_WRITE) 17822903ff01SAl Viro error = do_tee(in.file, out.file, 17832903ff01SAl Viro len, flags); 17842903ff01SAl Viro fdput(out); 178570524490SJens Axboe } 178670524490SJens Axboe } 17872903ff01SAl Viro fdput(in); 178870524490SJens Axboe } 178970524490SJens Axboe 179070524490SJens Axboe return error; 179170524490SJens Axboe } 1792