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 */ 20be297968SChristoph Hellwig #include <linux/bvec.h> 215274f052SJens Axboe #include <linux/fs.h> 225274f052SJens Axboe #include <linux/file.h> 235274f052SJens Axboe #include <linux/pagemap.h> 24d6b29d7cSJens Axboe #include <linux/splice.h> 2508e552c6SKAMEZAWA Hiroyuki #include <linux/memcontrol.h> 265274f052SJens Axboe #include <linux/mm_inline.h> 275abc97aaSJens Axboe #include <linux/swap.h> 284f6f0bd2SJens Axboe #include <linux/writeback.h> 29630d9c47SPaul Gortmaker #include <linux/export.h> 304f6f0bd2SJens Axboe #include <linux/syscalls.h> 31912d35f8SJens Axboe #include <linux/uio.h> 3229ce2058SJames Morris #include <linux/security.h> 335a0e3ad6STejun Heo #include <linux/gfp.h> 3435f9c09fSEric Dumazet #include <linux/socket.h> 3576b021d0SAl Viro #include <linux/compat.h> 36174cd4b1SIngo Molnar #include <linux/sched/signal.h> 37174cd4b1SIngo Molnar 3806ae43f3SAl Viro #include "internal.h" 395274f052SJens Axboe 4083f9135bSJens Axboe /* 4183f9135bSJens Axboe * Attempt to steal a page from a pipe buffer. This should perhaps go into 4283f9135bSJens Axboe * a vm helper function, it's already simplified quite a bit by the 4383f9135bSJens Axboe * addition of remove_mapping(). If success is returned, the caller may 4483f9135bSJens Axboe * attempt to reuse this page for another destination. 4583f9135bSJens Axboe */ 4676ad4d11SJens Axboe static int page_cache_pipe_buf_steal(struct pipe_inode_info *pipe, 475abc97aaSJens Axboe struct pipe_buffer *buf) 485abc97aaSJens Axboe { 495abc97aaSJens Axboe struct page *page = buf->page; 509e94cd4fSJens Axboe struct address_space *mapping; 515abc97aaSJens Axboe 529e0267c2SJens Axboe lock_page(page); 539e0267c2SJens Axboe 549e94cd4fSJens Axboe mapping = page_mapping(page); 559e94cd4fSJens Axboe if (mapping) { 565abc97aaSJens Axboe WARN_ON(!PageUptodate(page)); 575abc97aaSJens Axboe 58ad8d6f0aSJens Axboe /* 599e94cd4fSJens Axboe * At least for ext2 with nobh option, we need to wait on 609e94cd4fSJens Axboe * writeback completing on this page, since we'll remove it 619e94cd4fSJens Axboe * from the pagecache. Otherwise truncate wont wait on the 629e94cd4fSJens Axboe * page, allowing the disk blocks to be reused by someone else 639e94cd4fSJens Axboe * before we actually wrote our data to them. fs corruption 649e94cd4fSJens Axboe * ensues. 65ad8d6f0aSJens Axboe */ 66ad8d6f0aSJens Axboe wait_on_page_writeback(page); 67ad8d6f0aSJens Axboe 68266cf658SDavid Howells if (page_has_private(page) && 69266cf658SDavid Howells !try_to_release_page(page, GFP_KERNEL)) 70ca39d651SJens Axboe goto out_unlock; 714f6f0bd2SJens Axboe 729e94cd4fSJens Axboe /* 739e94cd4fSJens Axboe * If we succeeded in removing the mapping, set LRU flag 749e94cd4fSJens Axboe * and return good. 759e94cd4fSJens Axboe */ 769e94cd4fSJens Axboe if (remove_mapping(mapping, page)) { 771432873aSJens Axboe buf->flags |= PIPE_BUF_FLAG_LRU; 785abc97aaSJens Axboe return 0; 795abc97aaSJens Axboe } 809e94cd4fSJens Axboe } 819e94cd4fSJens Axboe 829e94cd4fSJens Axboe /* 839e94cd4fSJens Axboe * Raced with truncate or failed to remove page from current 849e94cd4fSJens Axboe * address space, unlock and return failure. 859e94cd4fSJens Axboe */ 86ca39d651SJens Axboe out_unlock: 879e94cd4fSJens Axboe unlock_page(page); 889e94cd4fSJens Axboe return 1; 899e94cd4fSJens Axboe } 905abc97aaSJens Axboe 9176ad4d11SJens Axboe static void page_cache_pipe_buf_release(struct pipe_inode_info *pipe, 925274f052SJens Axboe struct pipe_buffer *buf) 935274f052SJens Axboe { 9409cbfeafSKirill A. Shutemov put_page(buf->page); 951432873aSJens Axboe buf->flags &= ~PIPE_BUF_FLAG_LRU; 965274f052SJens Axboe } 975274f052SJens Axboe 980845718dSJens Axboe /* 990845718dSJens Axboe * Check whether the contents of buf is OK to access. Since the content 1000845718dSJens Axboe * is a page cache page, IO may be in flight. 1010845718dSJens Axboe */ 102cac36bb0SJens Axboe static int page_cache_pipe_buf_confirm(struct pipe_inode_info *pipe, 1035274f052SJens Axboe struct pipe_buffer *buf) 1045274f052SJens Axboe { 1055274f052SJens Axboe struct page *page = buf->page; 10649d0b21bSJens Axboe int err; 1075274f052SJens Axboe 1085274f052SJens Axboe if (!PageUptodate(page)) { 10949d0b21bSJens Axboe lock_page(page); 1105274f052SJens Axboe 11149d0b21bSJens Axboe /* 11249d0b21bSJens Axboe * Page got truncated/unhashed. This will cause a 0-byte 11373d62d83SIngo Molnar * splice, if this is the first page. 11449d0b21bSJens Axboe */ 1155274f052SJens Axboe if (!page->mapping) { 11649d0b21bSJens Axboe err = -ENODATA; 11749d0b21bSJens Axboe goto error; 1185274f052SJens Axboe } 1195274f052SJens Axboe 12049d0b21bSJens Axboe /* 12173d62d83SIngo Molnar * Uh oh, read-error from disk. 12249d0b21bSJens Axboe */ 12349d0b21bSJens Axboe if (!PageUptodate(page)) { 12449d0b21bSJens Axboe err = -EIO; 12549d0b21bSJens Axboe goto error; 12649d0b21bSJens Axboe } 12749d0b21bSJens Axboe 12849d0b21bSJens Axboe /* 129f84d7519SJens Axboe * Page is ok afterall, we are done. 13049d0b21bSJens Axboe */ 13149d0b21bSJens Axboe unlock_page(page); 13249d0b21bSJens Axboe } 13349d0b21bSJens Axboe 134f84d7519SJens Axboe return 0; 13549d0b21bSJens Axboe error: 13649d0b21bSJens Axboe unlock_page(page); 137f84d7519SJens Axboe return err; 13870524490SJens Axboe } 13970524490SJens Axboe 140708e3508SHugh Dickins const struct pipe_buf_operations page_cache_pipe_buf_ops = { 1415274f052SJens Axboe .can_merge = 0, 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 = { 159912d35f8SJens Axboe .can_merge = 0, 160cac36bb0SJens Axboe .confirm = generic_pipe_buf_confirm, 161912d35f8SJens Axboe .release = page_cache_pipe_buf_release, 162912d35f8SJens Axboe .steal = user_page_pipe_buf_steal, 163f84d7519SJens Axboe .get = generic_pipe_buf_get, 164912d35f8SJens Axboe }; 165912d35f8SJens Axboe 166825cdcb1SNamhyung Kim static void wakeup_pipe_readers(struct pipe_inode_info *pipe) 167825cdcb1SNamhyung Kim { 168825cdcb1SNamhyung Kim smp_mb(); 169825cdcb1SNamhyung Kim if (waitqueue_active(&pipe->wait)) 170825cdcb1SNamhyung Kim wake_up_interruptible(&pipe->wait); 171825cdcb1SNamhyung Kim kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN); 172825cdcb1SNamhyung Kim } 173825cdcb1SNamhyung Kim 174932cc6d4SJens Axboe /** 175932cc6d4SJens Axboe * splice_to_pipe - fill passed data into a pipe 176932cc6d4SJens Axboe * @pipe: pipe to fill 177932cc6d4SJens Axboe * @spd: data to fill 178932cc6d4SJens Axboe * 179932cc6d4SJens Axboe * Description: 18079685b8dSRandy Dunlap * @spd contains a map of pages and len/offset tuples, along with 181932cc6d4SJens Axboe * the struct pipe_buf_operations associated with these pages. This 182932cc6d4SJens Axboe * function will link that data to the pipe. 183932cc6d4SJens Axboe * 18483f9135bSJens Axboe */ 185d6b29d7cSJens Axboe ssize_t splice_to_pipe(struct pipe_inode_info *pipe, 186912d35f8SJens Axboe struct splice_pipe_desc *spd) 1875274f052SJens Axboe { 18800de00bdSJens Axboe unsigned int spd_pages = spd->nr_pages; 1898924feffSAl Viro int ret = 0, page_nr = 0; 1905274f052SJens Axboe 191d6785d91SRabin Vincent if (!spd_pages) 192d6785d91SRabin Vincent return 0; 193d6785d91SRabin Vincent 1948924feffSAl Viro if (unlikely(!pipe->readers)) { 1955274f052SJens Axboe send_sig(SIGPIPE, current, 0); 1965274f052SJens Axboe ret = -EPIPE; 1978924feffSAl Viro goto out; 1985274f052SJens Axboe } 1995274f052SJens Axboe 2008924feffSAl Viro while (pipe->nrbufs < pipe->buffers) { 20135f3d14dSJens Axboe int newbuf = (pipe->curbuf + pipe->nrbufs) & (pipe->buffers - 1); 2023a326a2cSIngo Molnar struct pipe_buffer *buf = pipe->bufs + newbuf; 2035274f052SJens Axboe 204912d35f8SJens Axboe buf->page = spd->pages[page_nr]; 205912d35f8SJens Axboe buf->offset = spd->partial[page_nr].offset; 206912d35f8SJens Axboe buf->len = spd->partial[page_nr].len; 207497f9625SJens Axboe buf->private = spd->partial[page_nr].private; 208912d35f8SJens Axboe buf->ops = spd->ops; 2095a81e6a1SMiklos Szeredi buf->flags = 0; 2107afa6fd0SJens Axboe 2116f767b04SJens Axboe pipe->nrbufs++; 212912d35f8SJens Axboe page_nr++; 213912d35f8SJens Axboe ret += buf->len; 214912d35f8SJens Axboe 215912d35f8SJens Axboe if (!--spd->nr_pages) 2165274f052SJens Axboe break; 2175274f052SJens Axboe } 2185274f052SJens Axboe 21929e35094SLinus Torvalds if (!ret) 22029e35094SLinus Torvalds ret = -EAGAIN; 22129e35094SLinus Torvalds 2228924feffSAl Viro out: 22300de00bdSJens Axboe while (page_nr < spd_pages) 224bbdfc2f7SJens Axboe spd->spd_release(spd, page_nr++); 2255274f052SJens Axboe 2265274f052SJens Axboe return ret; 2275274f052SJens Axboe } 2282b514574SHannes Frederic Sowa EXPORT_SYMBOL_GPL(splice_to_pipe); 2295274f052SJens Axboe 23079fddc4eSAl Viro ssize_t add_to_pipe(struct pipe_inode_info *pipe, struct pipe_buffer *buf) 23179fddc4eSAl Viro { 23279fddc4eSAl Viro int ret; 23379fddc4eSAl Viro 23479fddc4eSAl Viro if (unlikely(!pipe->readers)) { 23579fddc4eSAl Viro send_sig(SIGPIPE, current, 0); 23679fddc4eSAl Viro ret = -EPIPE; 23779fddc4eSAl Viro } else if (pipe->nrbufs == pipe->buffers) { 23879fddc4eSAl Viro ret = -EAGAIN; 23979fddc4eSAl Viro } else { 24079fddc4eSAl Viro int newbuf = (pipe->curbuf + pipe->nrbufs) & (pipe->buffers - 1); 24179fddc4eSAl Viro pipe->bufs[newbuf] = *buf; 24279fddc4eSAl Viro pipe->nrbufs++; 24379fddc4eSAl Viro return buf->len; 24479fddc4eSAl Viro } 245a779638cSMiklos Szeredi pipe_buf_release(pipe, buf); 24679fddc4eSAl Viro return ret; 24779fddc4eSAl Viro } 24879fddc4eSAl Viro EXPORT_SYMBOL(add_to_pipe); 24979fddc4eSAl Viro 25035f3d14dSJens Axboe /* 25135f3d14dSJens Axboe * Check if we need to grow the arrays holding pages and partial page 25235f3d14dSJens Axboe * descriptions. 25335f3d14dSJens Axboe */ 254047fe360SEric Dumazet int splice_grow_spd(const struct pipe_inode_info *pipe, struct splice_pipe_desc *spd) 25535f3d14dSJens Axboe { 2566aa7de05SMark Rutland unsigned int buffers = READ_ONCE(pipe->buffers); 257047fe360SEric Dumazet 258047fe360SEric Dumazet spd->nr_pages_max = buffers; 259047fe360SEric Dumazet if (buffers <= PIPE_DEF_BUFFERS) 26035f3d14dSJens Axboe return 0; 26135f3d14dSJens Axboe 262047fe360SEric Dumazet spd->pages = kmalloc(buffers * sizeof(struct page *), GFP_KERNEL); 263047fe360SEric Dumazet spd->partial = kmalloc(buffers * sizeof(struct partial_page), 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 30382c156f8SAl Viro iov_iter_pipe(&to, ITER_PIPE | 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 .can_merge = 0, 3296818173bSMiklos Szeredi .confirm = generic_pipe_buf_confirm, 3306818173bSMiklos Szeredi .release = generic_pipe_buf_release, 3316818173bSMiklos Szeredi .steal = generic_pipe_buf_steal, 3326818173bSMiklos Szeredi .get = generic_pipe_buf_get, 3336818173bSMiklos Szeredi }; 3346818173bSMiklos Szeredi 33528a625cbSMiklos Szeredi static int generic_pipe_buf_nosteal(struct pipe_inode_info *pipe, 33628a625cbSMiklos Szeredi struct pipe_buffer *buf) 33728a625cbSMiklos Szeredi { 33828a625cbSMiklos Szeredi return 1; 33928a625cbSMiklos Szeredi } 34028a625cbSMiklos Szeredi 34128a625cbSMiklos Szeredi /* Pipe buffer operations for a socket and similar. */ 34228a625cbSMiklos Szeredi const struct pipe_buf_operations nosteal_pipe_buf_ops = { 34328a625cbSMiklos Szeredi .can_merge = 0, 34428a625cbSMiklos Szeredi .confirm = generic_pipe_buf_confirm, 34528a625cbSMiklos Szeredi .release = generic_pipe_buf_release, 34628a625cbSMiklos Szeredi .steal = generic_pipe_buf_nosteal, 34728a625cbSMiklos Szeredi .get = generic_pipe_buf_get, 34828a625cbSMiklos Szeredi }; 34928a625cbSMiklos Szeredi EXPORT_SYMBOL(nosteal_pipe_buf_ops); 35028a625cbSMiklos Szeredi 351523ac9afSAl Viro static ssize_t kernel_readv(struct file *file, const struct kvec *vec, 3526818173bSMiklos Szeredi unsigned long vlen, loff_t offset) 3536818173bSMiklos Szeredi { 3546818173bSMiklos Szeredi mm_segment_t old_fs; 3556818173bSMiklos Szeredi loff_t pos = offset; 3566818173bSMiklos Szeredi ssize_t res; 3576818173bSMiklos Szeredi 3586818173bSMiklos Szeredi old_fs = get_fs(); 3596818173bSMiklos Szeredi set_fs(get_ds()); 3606818173bSMiklos Szeredi /* The cast to a user pointer is valid due to the set_fs() */ 361793b80efSChristoph Hellwig res = vfs_readv(file, (const struct iovec __user *)vec, vlen, &pos, 0); 3626818173bSMiklos Szeredi set_fs(old_fs); 3636818173bSMiklos Szeredi 3646818173bSMiklos Szeredi return res; 3656818173bSMiklos Szeredi } 3666818173bSMiklos Szeredi 36782c156f8SAl Viro static ssize_t default_file_splice_read(struct file *in, loff_t *ppos, 3686818173bSMiklos Szeredi struct pipe_inode_info *pipe, size_t len, 3696818173bSMiklos Szeredi unsigned int flags) 3706818173bSMiklos Szeredi { 371523ac9afSAl Viro struct kvec *vec, __vec[PIPE_DEF_BUFFERS]; 372523ac9afSAl Viro struct iov_iter to; 373523ac9afSAl Viro struct page **pages; 3746818173bSMiklos Szeredi unsigned int nr_pages; 37513c0f52bSAl Viro size_t offset, base, copied = 0; 3766818173bSMiklos Szeredi ssize_t res; 3776818173bSMiklos Szeredi int i; 3786818173bSMiklos Szeredi 379523ac9afSAl Viro if (pipe->nrbufs == pipe->buffers) 380523ac9afSAl Viro return -EAGAIN; 381523ac9afSAl Viro 382523ac9afSAl Viro /* 383523ac9afSAl Viro * Try to keep page boundaries matching to source pagecache ones - 384523ac9afSAl Viro * it probably won't be much help, but... 385523ac9afSAl Viro */ 386523ac9afSAl Viro offset = *ppos & ~PAGE_MASK; 387523ac9afSAl Viro 388523ac9afSAl Viro iov_iter_pipe(&to, ITER_PIPE | READ, pipe, len + offset); 389523ac9afSAl Viro 39013c0f52bSAl Viro res = iov_iter_get_pages_alloc(&to, &pages, len + offset, &base); 391523ac9afSAl Viro if (res <= 0) 39235f3d14dSJens Axboe return -ENOMEM; 39335f3d14dSJens Axboe 39413c0f52bSAl Viro nr_pages = DIV_ROUND_UP(res + base, PAGE_SIZE); 395523ac9afSAl Viro 39635f3d14dSJens Axboe vec = __vec; 397523ac9afSAl Viro if (nr_pages > PIPE_DEF_BUFFERS) { 398523ac9afSAl Viro vec = kmalloc(nr_pages * sizeof(struct kvec), GFP_KERNEL); 399523ac9afSAl Viro if (unlikely(!vec)) { 400523ac9afSAl Viro res = -ENOMEM; 401523ac9afSAl Viro goto out; 402523ac9afSAl Viro } 40335f3d14dSJens Axboe } 40435f3d14dSJens Axboe 405523ac9afSAl Viro pipe->bufs[to.idx].offset = offset; 406523ac9afSAl Viro pipe->bufs[to.idx].len -= offset; 4076818173bSMiklos Szeredi 408523ac9afSAl Viro for (i = 0; i < nr_pages; i++) { 409523ac9afSAl Viro size_t this_len = min_t(size_t, len, PAGE_SIZE - offset); 410523ac9afSAl Viro vec[i].iov_base = page_address(pages[i]) + offset; 4116818173bSMiklos Szeredi vec[i].iov_len = this_len; 4126818173bSMiklos Szeredi len -= this_len; 4136818173bSMiklos Szeredi offset = 0; 4146818173bSMiklos Szeredi } 4156818173bSMiklos Szeredi 416523ac9afSAl Viro res = kernel_readv(in, vec, nr_pages, *ppos); 417523ac9afSAl Viro if (res > 0) { 418523ac9afSAl Viro copied = res; 4196818173bSMiklos Szeredi *ppos += res; 420523ac9afSAl Viro } 4216818173bSMiklos Szeredi 42235f3d14dSJens Axboe if (vec != __vec) 42335f3d14dSJens Axboe kfree(vec); 424523ac9afSAl Viro out: 425523ac9afSAl Viro for (i = 0; i < nr_pages; i++) 426523ac9afSAl Viro put_page(pages[i]); 427523ac9afSAl Viro kvfree(pages); 428523ac9afSAl Viro iov_iter_advance(&to, copied); /* truncates and discards */ 4296818173bSMiklos Szeredi return res; 4306818173bSMiklos Szeredi } 4316818173bSMiklos Szeredi 4325274f052SJens Axboe /* 4334f6f0bd2SJens Axboe * Send 'sd->len' bytes to socket from 'sd->file' at position 'sd->pos' 434016b661eSJens Axboe * using sendpage(). Return the number of bytes sent. 4355274f052SJens Axboe */ 43676ad4d11SJens Axboe static int pipe_to_sendpage(struct pipe_inode_info *pipe, 4375274f052SJens Axboe struct pipe_buffer *buf, struct splice_desc *sd) 4385274f052SJens Axboe { 4396a14b90bSJens Axboe struct file *file = sd->u.file; 4405274f052SJens Axboe loff_t pos = sd->pos; 441a8adbe37SMichał Mirosław int more; 4425274f052SJens Axboe 44372c2d531SAl Viro if (!likely(file->f_op->sendpage)) 444a8adbe37SMichał Mirosław return -EINVAL; 445a8adbe37SMichał Mirosław 44635f9c09fSEric Dumazet more = (sd->flags & SPLICE_F_MORE) ? MSG_MORE : 0; 447ae62ca7bSEric Dumazet 448ae62ca7bSEric Dumazet if (sd->len < sd->total_len && pipe->nrbufs > 1) 44935f9c09fSEric Dumazet more |= MSG_SENDPAGE_NOTLAST; 450ae62ca7bSEric Dumazet 451a8adbe37SMichał Mirosław return file->f_op->sendpage(file, buf->page, buf->offset, 452f84d7519SJens Axboe sd->len, &pos, more); 4535274f052SJens Axboe } 4545274f052SJens Axboe 455b3c2d2ddSMiklos Szeredi static void wakeup_pipe_writers(struct pipe_inode_info *pipe) 456b3c2d2ddSMiklos Szeredi { 457b3c2d2ddSMiklos Szeredi smp_mb(); 458b3c2d2ddSMiklos Szeredi if (waitqueue_active(&pipe->wait)) 459b3c2d2ddSMiklos Szeredi wake_up_interruptible(&pipe->wait); 460b3c2d2ddSMiklos Szeredi kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT); 461b3c2d2ddSMiklos Szeredi } 462b3c2d2ddSMiklos Szeredi 463b3c2d2ddSMiklos Szeredi /** 464b3c2d2ddSMiklos Szeredi * splice_from_pipe_feed - feed available data from a pipe to a file 465b3c2d2ddSMiklos Szeredi * @pipe: pipe to splice from 466b3c2d2ddSMiklos Szeredi * @sd: information to @actor 467b3c2d2ddSMiklos Szeredi * @actor: handler that splices the data 468b3c2d2ddSMiklos Szeredi * 469b3c2d2ddSMiklos Szeredi * Description: 470b3c2d2ddSMiklos Szeredi * This function loops over the pipe and calls @actor to do the 471b3c2d2ddSMiklos Szeredi * actual moving of a single struct pipe_buffer to the desired 472b3c2d2ddSMiklos Szeredi * destination. It returns when there's no more buffers left in 473b3c2d2ddSMiklos Szeredi * the pipe or if the requested number of bytes (@sd->total_len) 474b3c2d2ddSMiklos Szeredi * have been copied. It returns a positive number (one) if the 475b3c2d2ddSMiklos Szeredi * pipe needs to be filled with more data, zero if the required 476b3c2d2ddSMiklos Szeredi * number of bytes have been copied and -errno on error. 477b3c2d2ddSMiklos Szeredi * 478b3c2d2ddSMiklos Szeredi * This, together with splice_from_pipe_{begin,end,next}, may be 479b3c2d2ddSMiklos Szeredi * used to implement the functionality of __splice_from_pipe() when 480b3c2d2ddSMiklos Szeredi * locking is required around copying the pipe buffers to the 481b3c2d2ddSMiklos Szeredi * destination. 482b3c2d2ddSMiklos Szeredi */ 48396f9bc8fSAl Viro static int splice_from_pipe_feed(struct pipe_inode_info *pipe, struct splice_desc *sd, 484b3c2d2ddSMiklos Szeredi splice_actor *actor) 485b3c2d2ddSMiklos Szeredi { 486b3c2d2ddSMiklos Szeredi int ret; 487b3c2d2ddSMiklos Szeredi 488b3c2d2ddSMiklos Szeredi while (pipe->nrbufs) { 489b3c2d2ddSMiklos Szeredi struct pipe_buffer *buf = pipe->bufs + pipe->curbuf; 490b3c2d2ddSMiklos Szeredi 491b3c2d2ddSMiklos Szeredi sd->len = buf->len; 492b3c2d2ddSMiklos Szeredi if (sd->len > sd->total_len) 493b3c2d2ddSMiklos Szeredi sd->len = sd->total_len; 494b3c2d2ddSMiklos Szeredi 495fba597dbSMiklos Szeredi ret = pipe_buf_confirm(pipe, buf); 496a8adbe37SMichał Mirosław if (unlikely(ret)) { 497b3c2d2ddSMiklos Szeredi if (ret == -ENODATA) 498b3c2d2ddSMiklos Szeredi ret = 0; 499b3c2d2ddSMiklos Szeredi return ret; 500b3c2d2ddSMiklos Szeredi } 501a8adbe37SMichał Mirosław 502a8adbe37SMichał Mirosław ret = actor(pipe, buf, sd); 503a8adbe37SMichał Mirosław if (ret <= 0) 504a8adbe37SMichał Mirosław return ret; 505a8adbe37SMichał Mirosław 506b3c2d2ddSMiklos Szeredi buf->offset += ret; 507b3c2d2ddSMiklos Szeredi buf->len -= ret; 508b3c2d2ddSMiklos Szeredi 509b3c2d2ddSMiklos Szeredi sd->num_spliced += ret; 510b3c2d2ddSMiklos Szeredi sd->len -= ret; 511b3c2d2ddSMiklos Szeredi sd->pos += ret; 512b3c2d2ddSMiklos Szeredi sd->total_len -= ret; 513b3c2d2ddSMiklos Szeredi 514b3c2d2ddSMiklos Szeredi if (!buf->len) { 515a779638cSMiklos Szeredi pipe_buf_release(pipe, buf); 51635f3d14dSJens Axboe pipe->curbuf = (pipe->curbuf + 1) & (pipe->buffers - 1); 517b3c2d2ddSMiklos Szeredi pipe->nrbufs--; 5186447a3cfSAl Viro if (pipe->files) 519b3c2d2ddSMiklos Szeredi sd->need_wakeup = true; 520b3c2d2ddSMiklos Szeredi } 521b3c2d2ddSMiklos Szeredi 522b3c2d2ddSMiklos Szeredi if (!sd->total_len) 523b3c2d2ddSMiklos Szeredi return 0; 524b3c2d2ddSMiklos Szeredi } 525b3c2d2ddSMiklos Szeredi 526b3c2d2ddSMiklos Szeredi return 1; 527b3c2d2ddSMiklos Szeredi } 528b3c2d2ddSMiklos Szeredi 529b3c2d2ddSMiklos Szeredi /** 530b3c2d2ddSMiklos Szeredi * splice_from_pipe_next - wait for some data to splice from 531b3c2d2ddSMiklos Szeredi * @pipe: pipe to splice from 532b3c2d2ddSMiklos Szeredi * @sd: information about the splice operation 533b3c2d2ddSMiklos Szeredi * 534b3c2d2ddSMiklos Szeredi * Description: 535b3c2d2ddSMiklos Szeredi * This function will wait for some data and return a positive 536b3c2d2ddSMiklos Szeredi * value (one) if pipe buffers are available. It will return zero 537b3c2d2ddSMiklos Szeredi * or -errno if no more data needs to be spliced. 538b3c2d2ddSMiklos Szeredi */ 53996f9bc8fSAl Viro static int splice_from_pipe_next(struct pipe_inode_info *pipe, struct splice_desc *sd) 540b3c2d2ddSMiklos Szeredi { 541c725bfceSJan Kara /* 542c725bfceSJan Kara * Check for signal early to make process killable when there are 543c725bfceSJan Kara * always buffers available 544c725bfceSJan Kara */ 545c725bfceSJan Kara if (signal_pending(current)) 546c725bfceSJan Kara return -ERESTARTSYS; 547c725bfceSJan Kara 548b3c2d2ddSMiklos Szeredi while (!pipe->nrbufs) { 549b3c2d2ddSMiklos Szeredi if (!pipe->writers) 550b3c2d2ddSMiklos Szeredi return 0; 551b3c2d2ddSMiklos Szeredi 552b3c2d2ddSMiklos Szeredi if (!pipe->waiting_writers && sd->num_spliced) 553b3c2d2ddSMiklos Szeredi return 0; 554b3c2d2ddSMiklos Szeredi 555b3c2d2ddSMiklos Szeredi if (sd->flags & SPLICE_F_NONBLOCK) 556b3c2d2ddSMiklos Szeredi return -EAGAIN; 557b3c2d2ddSMiklos Szeredi 558b3c2d2ddSMiklos Szeredi if (signal_pending(current)) 559b3c2d2ddSMiklos Szeredi return -ERESTARTSYS; 560b3c2d2ddSMiklos Szeredi 561b3c2d2ddSMiklos Szeredi if (sd->need_wakeup) { 562b3c2d2ddSMiklos Szeredi wakeup_pipe_writers(pipe); 563b3c2d2ddSMiklos Szeredi sd->need_wakeup = false; 564b3c2d2ddSMiklos Szeredi } 565b3c2d2ddSMiklos Szeredi 566b3c2d2ddSMiklos Szeredi pipe_wait(pipe); 567b3c2d2ddSMiklos Szeredi } 568b3c2d2ddSMiklos Szeredi 569b3c2d2ddSMiklos Szeredi return 1; 570b3c2d2ddSMiklos Szeredi } 571b3c2d2ddSMiklos Szeredi 572b3c2d2ddSMiklos Szeredi /** 573b3c2d2ddSMiklos Szeredi * splice_from_pipe_begin - start splicing from pipe 574b80901bbSRandy Dunlap * @sd: information about the splice operation 575b3c2d2ddSMiklos Szeredi * 576b3c2d2ddSMiklos Szeredi * Description: 577b3c2d2ddSMiklos Szeredi * This function should be called before a loop containing 578b3c2d2ddSMiklos Szeredi * splice_from_pipe_next() and splice_from_pipe_feed() to 579b3c2d2ddSMiklos Szeredi * initialize the necessary fields of @sd. 580b3c2d2ddSMiklos Szeredi */ 58196f9bc8fSAl Viro static void splice_from_pipe_begin(struct splice_desc *sd) 582b3c2d2ddSMiklos Szeredi { 583b3c2d2ddSMiklos Szeredi sd->num_spliced = 0; 584b3c2d2ddSMiklos Szeredi sd->need_wakeup = false; 585b3c2d2ddSMiklos Szeredi } 586b3c2d2ddSMiklos Szeredi 587b3c2d2ddSMiklos Szeredi /** 588b3c2d2ddSMiklos Szeredi * splice_from_pipe_end - finish splicing from pipe 589b3c2d2ddSMiklos Szeredi * @pipe: pipe to splice from 590b3c2d2ddSMiklos Szeredi * @sd: information about the splice operation 591b3c2d2ddSMiklos Szeredi * 592b3c2d2ddSMiklos Szeredi * Description: 593b3c2d2ddSMiklos Szeredi * This function will wake up pipe writers if necessary. It should 594b3c2d2ddSMiklos Szeredi * be called after a loop containing splice_from_pipe_next() and 595b3c2d2ddSMiklos Szeredi * splice_from_pipe_feed(). 596b3c2d2ddSMiklos Szeredi */ 59796f9bc8fSAl Viro static void splice_from_pipe_end(struct pipe_inode_info *pipe, struct splice_desc *sd) 598b3c2d2ddSMiklos Szeredi { 599b3c2d2ddSMiklos Szeredi if (sd->need_wakeup) 600b3c2d2ddSMiklos Szeredi wakeup_pipe_writers(pipe); 601b3c2d2ddSMiklos Szeredi } 602b3c2d2ddSMiklos Szeredi 603932cc6d4SJens Axboe /** 604932cc6d4SJens Axboe * __splice_from_pipe - splice data from a pipe to given actor 605932cc6d4SJens Axboe * @pipe: pipe to splice from 606932cc6d4SJens Axboe * @sd: information to @actor 607932cc6d4SJens Axboe * @actor: handler that splices the data 608932cc6d4SJens Axboe * 609932cc6d4SJens Axboe * Description: 610932cc6d4SJens Axboe * This function does little more than loop over the pipe and call 611932cc6d4SJens Axboe * @actor to do the actual moving of a single struct pipe_buffer to 612932cc6d4SJens Axboe * the desired destination. See pipe_to_file, pipe_to_sendpage, or 613932cc6d4SJens Axboe * pipe_to_user. 614932cc6d4SJens Axboe * 61583f9135bSJens Axboe */ 616c66ab6faSJens Axboe ssize_t __splice_from_pipe(struct pipe_inode_info *pipe, struct splice_desc *sd, 617c66ab6faSJens Axboe splice_actor *actor) 6185274f052SJens Axboe { 619b3c2d2ddSMiklos Szeredi int ret; 6205274f052SJens Axboe 621b3c2d2ddSMiklos Szeredi splice_from_pipe_begin(sd); 622b3c2d2ddSMiklos Szeredi do { 623c2489e07SJan Kara cond_resched(); 624b3c2d2ddSMiklos Szeredi ret = splice_from_pipe_next(pipe, sd); 625b3c2d2ddSMiklos Szeredi if (ret > 0) 626b3c2d2ddSMiklos Szeredi ret = splice_from_pipe_feed(pipe, sd, actor); 627b3c2d2ddSMiklos Szeredi } while (ret > 0); 628b3c2d2ddSMiklos Szeredi splice_from_pipe_end(pipe, sd); 6295274f052SJens Axboe 630b3c2d2ddSMiklos Szeredi return sd->num_spliced ? sd->num_spliced : ret; 6315274f052SJens Axboe } 63240bee44eSMark Fasheh EXPORT_SYMBOL(__splice_from_pipe); 6335274f052SJens Axboe 634932cc6d4SJens Axboe /** 635932cc6d4SJens Axboe * splice_from_pipe - splice data from a pipe to a file 636932cc6d4SJens Axboe * @pipe: pipe to splice from 637932cc6d4SJens Axboe * @out: file to splice to 638932cc6d4SJens Axboe * @ppos: position in @out 639932cc6d4SJens Axboe * @len: how many bytes to splice 640932cc6d4SJens Axboe * @flags: splice modifier flags 641932cc6d4SJens Axboe * @actor: handler that splices the data 642932cc6d4SJens Axboe * 643932cc6d4SJens Axboe * Description: 6442933970bSMiklos Szeredi * See __splice_from_pipe. This function locks the pipe inode, 645932cc6d4SJens Axboe * otherwise it's identical to __splice_from_pipe(). 646932cc6d4SJens Axboe * 647932cc6d4SJens Axboe */ 6486da61809SMark Fasheh ssize_t splice_from_pipe(struct pipe_inode_info *pipe, struct file *out, 6496da61809SMark Fasheh loff_t *ppos, size_t len, unsigned int flags, 6506da61809SMark Fasheh splice_actor *actor) 6516da61809SMark Fasheh { 6526da61809SMark Fasheh ssize_t ret; 653c66ab6faSJens Axboe struct splice_desc sd = { 654c66ab6faSJens Axboe .total_len = len, 655c66ab6faSJens Axboe .flags = flags, 656c66ab6faSJens Axboe .pos = *ppos, 6576a14b90bSJens Axboe .u.file = out, 658c66ab6faSJens Axboe }; 6596da61809SMark Fasheh 66061e0d47cSMiklos Szeredi pipe_lock(pipe); 661c66ab6faSJens Axboe ret = __splice_from_pipe(pipe, &sd, actor); 66261e0d47cSMiklos Szeredi pipe_unlock(pipe); 6636da61809SMark Fasheh 6646da61809SMark Fasheh return ret; 6656da61809SMark Fasheh } 6666da61809SMark Fasheh 6676da61809SMark Fasheh /** 6688d020765SAl Viro * iter_file_splice_write - splice data from a pipe to a file 6698d020765SAl Viro * @pipe: pipe info 6708d020765SAl Viro * @out: file to write to 6718d020765SAl Viro * @ppos: position in @out 6728d020765SAl Viro * @len: number of bytes to splice 6738d020765SAl Viro * @flags: splice modifier flags 6748d020765SAl Viro * 6758d020765SAl Viro * Description: 6768d020765SAl Viro * Will either move or copy pages (determined by @flags options) from 6778d020765SAl Viro * the given pipe inode to the given file. 6788d020765SAl Viro * This one is ->write_iter-based. 6798d020765SAl Viro * 6808d020765SAl Viro */ 6818d020765SAl Viro ssize_t 6828d020765SAl Viro iter_file_splice_write(struct pipe_inode_info *pipe, struct file *out, 6838d020765SAl Viro loff_t *ppos, size_t len, unsigned int flags) 6848d020765SAl Viro { 6858d020765SAl Viro struct splice_desc sd = { 6868d020765SAl Viro .total_len = len, 6878d020765SAl Viro .flags = flags, 6888d020765SAl Viro .pos = *ppos, 6898d020765SAl Viro .u.file = out, 6908d020765SAl Viro }; 6918d020765SAl Viro int nbufs = pipe->buffers; 6928d020765SAl Viro struct bio_vec *array = kcalloc(nbufs, sizeof(struct bio_vec), 6938d020765SAl Viro GFP_KERNEL); 6948d020765SAl Viro ssize_t ret; 6958d020765SAl Viro 6968d020765SAl Viro if (unlikely(!array)) 6978d020765SAl Viro return -ENOMEM; 6988d020765SAl Viro 6998d020765SAl Viro pipe_lock(pipe); 7008d020765SAl Viro 7018d020765SAl Viro splice_from_pipe_begin(&sd); 7028d020765SAl Viro while (sd.total_len) { 7038d020765SAl Viro struct iov_iter from; 7048d020765SAl Viro size_t left; 7058d020765SAl Viro int n, idx; 7068d020765SAl Viro 7078d020765SAl Viro ret = splice_from_pipe_next(pipe, &sd); 7088d020765SAl Viro if (ret <= 0) 7098d020765SAl Viro break; 7108d020765SAl Viro 7118d020765SAl Viro if (unlikely(nbufs < pipe->buffers)) { 7128d020765SAl Viro kfree(array); 7138d020765SAl Viro nbufs = pipe->buffers; 7148d020765SAl Viro array = kcalloc(nbufs, sizeof(struct bio_vec), 7158d020765SAl Viro GFP_KERNEL); 7168d020765SAl Viro if (!array) { 7178d020765SAl Viro ret = -ENOMEM; 7188d020765SAl Viro break; 7198d020765SAl Viro } 7208d020765SAl Viro } 7218d020765SAl Viro 7228d020765SAl Viro /* build the vector */ 7238d020765SAl Viro left = sd.total_len; 7248d020765SAl Viro for (n = 0, idx = pipe->curbuf; left && n < pipe->nrbufs; n++, idx++) { 7258d020765SAl Viro struct pipe_buffer *buf = pipe->bufs + idx; 7268d020765SAl Viro size_t this_len = buf->len; 7278d020765SAl Viro 7288d020765SAl Viro if (this_len > left) 7298d020765SAl Viro this_len = left; 7308d020765SAl Viro 7318d020765SAl Viro if (idx == pipe->buffers - 1) 7328d020765SAl Viro idx = -1; 7338d020765SAl Viro 734fba597dbSMiklos Szeredi ret = pipe_buf_confirm(pipe, buf); 7358d020765SAl Viro if (unlikely(ret)) { 7368d020765SAl Viro if (ret == -ENODATA) 7378d020765SAl Viro ret = 0; 7388d020765SAl Viro goto done; 7398d020765SAl Viro } 7408d020765SAl Viro 7418d020765SAl Viro array[n].bv_page = buf->page; 7428d020765SAl Viro array[n].bv_len = this_len; 7438d020765SAl Viro array[n].bv_offset = buf->offset; 7448d020765SAl Viro left -= this_len; 7458d020765SAl Viro } 7468d020765SAl Viro 74705afcb77SAl Viro iov_iter_bvec(&from, ITER_BVEC | WRITE, array, n, 74805afcb77SAl Viro sd.total_len - left); 749abbb6589SChristoph Hellwig ret = vfs_iter_write(out, &from, &sd.pos, 0); 7508d020765SAl Viro if (ret <= 0) 7518d020765SAl Viro break; 7528d020765SAl Viro 7538d020765SAl Viro sd.num_spliced += ret; 7548d020765SAl Viro sd.total_len -= ret; 755dbe4e192SChristoph Hellwig *ppos = sd.pos; 7568d020765SAl Viro 7578d020765SAl Viro /* dismiss the fully eaten buffers, adjust the partial one */ 7588d020765SAl Viro while (ret) { 7598d020765SAl Viro struct pipe_buffer *buf = pipe->bufs + pipe->curbuf; 7608d020765SAl Viro if (ret >= buf->len) { 7618d020765SAl Viro ret -= buf->len; 7628d020765SAl Viro buf->len = 0; 763a779638cSMiklos Szeredi pipe_buf_release(pipe, buf); 7648d020765SAl Viro pipe->curbuf = (pipe->curbuf + 1) & (pipe->buffers - 1); 7658d020765SAl Viro pipe->nrbufs--; 7668d020765SAl Viro if (pipe->files) 7678d020765SAl Viro sd.need_wakeup = true; 7688d020765SAl Viro } else { 7698d020765SAl Viro buf->offset += ret; 7708d020765SAl Viro buf->len -= ret; 7718d020765SAl Viro ret = 0; 7728d020765SAl Viro } 7738d020765SAl Viro } 7748d020765SAl Viro } 7758d020765SAl Viro done: 7768d020765SAl Viro kfree(array); 7778d020765SAl Viro splice_from_pipe_end(pipe, &sd); 7788d020765SAl Viro 7798d020765SAl Viro pipe_unlock(pipe); 7808d020765SAl Viro 7818d020765SAl Viro if (sd.num_spliced) 7828d020765SAl Viro ret = sd.num_spliced; 7838d020765SAl Viro 7848d020765SAl Viro return ret; 7858d020765SAl Viro } 7868d020765SAl Viro 7878d020765SAl Viro EXPORT_SYMBOL(iter_file_splice_write); 7888d020765SAl Viro 789b2858d7dSMiklos Szeredi static int write_pipe_buf(struct pipe_inode_info *pipe, struct pipe_buffer *buf, 790b2858d7dSMiklos Szeredi struct splice_desc *sd) 7910b0a47f5SMiklos Szeredi { 792b2858d7dSMiklos Szeredi int ret; 793b2858d7dSMiklos Szeredi void *data; 79406ae43f3SAl Viro loff_t tmp = sd->pos; 795b2858d7dSMiklos Szeredi 796fbb32750SAl Viro data = kmap(buf->page); 79706ae43f3SAl Viro ret = __kernel_write(sd->u.file, data + buf->offset, sd->len, &tmp); 798fbb32750SAl Viro kunmap(buf->page); 799b2858d7dSMiklos Szeredi 800b2858d7dSMiklos Szeredi return ret; 8010b0a47f5SMiklos Szeredi } 8020b0a47f5SMiklos Szeredi 8030b0a47f5SMiklos Szeredi static ssize_t default_file_splice_write(struct pipe_inode_info *pipe, 8040b0a47f5SMiklos Szeredi struct file *out, loff_t *ppos, 8050b0a47f5SMiklos Szeredi size_t len, unsigned int flags) 8060b0a47f5SMiklos Szeredi { 807b2858d7dSMiklos Szeredi ssize_t ret; 8080b0a47f5SMiklos Szeredi 809b2858d7dSMiklos Szeredi ret = splice_from_pipe(pipe, out, ppos, len, flags, write_pipe_buf); 810b2858d7dSMiklos Szeredi if (ret > 0) 811b2858d7dSMiklos Szeredi *ppos += ret; 8120b0a47f5SMiklos Szeredi 813b2858d7dSMiklos Szeredi return ret; 8140b0a47f5SMiklos Szeredi } 8150b0a47f5SMiklos Szeredi 81683f9135bSJens Axboe /** 81783f9135bSJens Axboe * generic_splice_sendpage - splice data from a pipe to a socket 818932cc6d4SJens Axboe * @pipe: pipe to splice from 81983f9135bSJens Axboe * @out: socket to write to 820932cc6d4SJens Axboe * @ppos: position in @out 82183f9135bSJens Axboe * @len: number of bytes to splice 82283f9135bSJens Axboe * @flags: splice modifier flags 82383f9135bSJens Axboe * 824932cc6d4SJens Axboe * Description: 82583f9135bSJens Axboe * Will send @len bytes from the pipe to a network socket. No data copying 82683f9135bSJens Axboe * is involved. 82783f9135bSJens Axboe * 82883f9135bSJens Axboe */ 8293a326a2cSIngo Molnar ssize_t generic_splice_sendpage(struct pipe_inode_info *pipe, struct file *out, 830cbb7e577SJens Axboe loff_t *ppos, size_t len, unsigned int flags) 8315274f052SJens Axboe { 83200522fb4SJens Axboe return splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_sendpage); 8335274f052SJens Axboe } 8345274f052SJens Axboe 835059a8f37SJens Axboe EXPORT_SYMBOL(generic_splice_sendpage); 836a0f06780SJeff Garzik 83783f9135bSJens Axboe /* 83883f9135bSJens Axboe * Attempt to initiate a splice from pipe to file. 83983f9135bSJens Axboe */ 8403a326a2cSIngo Molnar static long do_splice_from(struct pipe_inode_info *pipe, struct file *out, 841cbb7e577SJens Axboe loff_t *ppos, size_t len, unsigned int flags) 8425274f052SJens Axboe { 8430b0a47f5SMiklos Szeredi ssize_t (*splice_write)(struct pipe_inode_info *, struct file *, 8440b0a47f5SMiklos Szeredi loff_t *, size_t, unsigned int); 8455274f052SJens Axboe 84672c2d531SAl Viro if (out->f_op->splice_write) 8470b0a47f5SMiklos Szeredi splice_write = out->f_op->splice_write; 848cc56f7deSChangli Gao else 8490b0a47f5SMiklos Szeredi splice_write = default_file_splice_write; 8500b0a47f5SMiklos Szeredi 851500368f7SAl Viro return splice_write(pipe, out, ppos, len, flags); 8525274f052SJens Axboe } 8535274f052SJens Axboe 85483f9135bSJens Axboe /* 85583f9135bSJens Axboe * Attempt to initiate a splice from a file to a pipe. 85683f9135bSJens Axboe */ 857cbb7e577SJens Axboe static long do_splice_to(struct file *in, loff_t *ppos, 858cbb7e577SJens Axboe struct pipe_inode_info *pipe, size_t len, 859cbb7e577SJens Axboe unsigned int flags) 8605274f052SJens Axboe { 8616818173bSMiklos Szeredi ssize_t (*splice_read)(struct file *, loff_t *, 8626818173bSMiklos Szeredi struct pipe_inode_info *, size_t, unsigned int); 8635274f052SJens Axboe int ret; 8645274f052SJens Axboe 86549570e9bSJens Axboe if (unlikely(!(in->f_mode & FMODE_READ))) 8665274f052SJens Axboe return -EBADF; 8675274f052SJens Axboe 868cbb7e577SJens Axboe ret = rw_verify_area(READ, in, ppos, len); 8695274f052SJens Axboe if (unlikely(ret < 0)) 8705274f052SJens Axboe return ret; 8715274f052SJens Axboe 87203cc0789SAl Viro if (unlikely(len > MAX_RW_COUNT)) 87303cc0789SAl Viro len = MAX_RW_COUNT; 87403cc0789SAl Viro 87572c2d531SAl Viro if (in->f_op->splice_read) 8766818173bSMiklos Szeredi splice_read = in->f_op->splice_read; 877cc56f7deSChangli Gao else 8786818173bSMiklos Szeredi splice_read = default_file_splice_read; 8796818173bSMiklos Szeredi 8806818173bSMiklos Szeredi return splice_read(in, ppos, pipe, len, flags); 8815274f052SJens Axboe } 8825274f052SJens Axboe 883932cc6d4SJens Axboe /** 884932cc6d4SJens Axboe * splice_direct_to_actor - splices data directly between two non-pipes 885932cc6d4SJens Axboe * @in: file to splice from 886932cc6d4SJens Axboe * @sd: actor information on where to splice to 887932cc6d4SJens Axboe * @actor: handles the data splicing 888932cc6d4SJens Axboe * 889932cc6d4SJens Axboe * Description: 890932cc6d4SJens Axboe * This is a special case helper to splice directly between two 891932cc6d4SJens Axboe * points, without requiring an explicit pipe. Internally an allocated 892932cc6d4SJens Axboe * pipe is cached in the process, and reused during the lifetime of 893932cc6d4SJens Axboe * that process. 894932cc6d4SJens Axboe * 895c66ab6faSJens Axboe */ 896c66ab6faSJens Axboe ssize_t splice_direct_to_actor(struct file *in, struct splice_desc *sd, 897c66ab6faSJens Axboe splice_direct_actor *actor) 898b92ce558SJens Axboe { 899b92ce558SJens Axboe struct pipe_inode_info *pipe; 900b92ce558SJens Axboe long ret, bytes; 901b92ce558SJens Axboe umode_t i_mode; 902c66ab6faSJens Axboe size_t len; 9030ff28d9fSChristophe Leroy int i, flags, more; 904b92ce558SJens Axboe 905b92ce558SJens Axboe /* 906b92ce558SJens Axboe * We require the input being a regular file, as we don't want to 907b92ce558SJens Axboe * randomly drop data for eg socket -> socket splicing. Use the 908b92ce558SJens Axboe * piped splicing for that! 909b92ce558SJens Axboe */ 910496ad9aaSAl Viro i_mode = file_inode(in)->i_mode; 911b92ce558SJens Axboe if (unlikely(!S_ISREG(i_mode) && !S_ISBLK(i_mode))) 912b92ce558SJens Axboe return -EINVAL; 913b92ce558SJens Axboe 914b92ce558SJens Axboe /* 915b92ce558SJens Axboe * neither in nor out is a pipe, setup an internal pipe attached to 916b92ce558SJens Axboe * 'out' and transfer the wanted data from 'in' to 'out' through that 917b92ce558SJens Axboe */ 918b92ce558SJens Axboe pipe = current->splice_pipe; 91949570e9bSJens Axboe if (unlikely(!pipe)) { 9207bee130eSAl Viro pipe = alloc_pipe_info(); 921b92ce558SJens Axboe if (!pipe) 922b92ce558SJens Axboe return -ENOMEM; 923b92ce558SJens Axboe 924b92ce558SJens Axboe /* 925b92ce558SJens Axboe * We don't have an immediate reader, but we'll read the stuff 92600522fb4SJens Axboe * out of the pipe right after the splice_to_pipe(). So set 927b92ce558SJens Axboe * PIPE_READERS appropriately. 928b92ce558SJens Axboe */ 929b92ce558SJens Axboe pipe->readers = 1; 930b92ce558SJens Axboe 931b92ce558SJens Axboe current->splice_pipe = pipe; 932b92ce558SJens Axboe } 933b92ce558SJens Axboe 934b92ce558SJens Axboe /* 93573d62d83SIngo Molnar * Do the splice. 936b92ce558SJens Axboe */ 937b92ce558SJens Axboe ret = 0; 938b92ce558SJens Axboe bytes = 0; 939c66ab6faSJens Axboe len = sd->total_len; 940c66ab6faSJens Axboe flags = sd->flags; 941c66ab6faSJens Axboe 942c66ab6faSJens Axboe /* 943c66ab6faSJens Axboe * Don't block on output, we have to drain the direct pipe. 944c66ab6faSJens Axboe */ 945c66ab6faSJens Axboe sd->flags &= ~SPLICE_F_NONBLOCK; 9460ff28d9fSChristophe Leroy more = sd->flags & SPLICE_F_MORE; 947b92ce558SJens Axboe 948b92ce558SJens Axboe while (len) { 94951a92c0fSJens Axboe size_t read_len; 950a82c53a0STom Zanussi loff_t pos = sd->pos, prev_pos = pos; 951b92ce558SJens Axboe 952bcd4f3acSJens Axboe ret = do_splice_to(in, &pos, pipe, len, flags); 95351a92c0fSJens Axboe if (unlikely(ret <= 0)) 954b92ce558SJens Axboe goto out_release; 955b92ce558SJens Axboe 956b92ce558SJens Axboe read_len = ret; 957c66ab6faSJens Axboe sd->total_len = read_len; 958b92ce558SJens Axboe 959b92ce558SJens Axboe /* 9600ff28d9fSChristophe Leroy * If more data is pending, set SPLICE_F_MORE 9610ff28d9fSChristophe Leroy * If this is the last data and SPLICE_F_MORE was not set 9620ff28d9fSChristophe Leroy * initially, clears it. 9630ff28d9fSChristophe Leroy */ 9640ff28d9fSChristophe Leroy if (read_len < len) 9650ff28d9fSChristophe Leroy sd->flags |= SPLICE_F_MORE; 9660ff28d9fSChristophe Leroy else if (!more) 9670ff28d9fSChristophe Leroy sd->flags &= ~SPLICE_F_MORE; 9680ff28d9fSChristophe Leroy /* 969b92ce558SJens Axboe * NOTE: nonblocking mode only applies to the input. We 970b92ce558SJens Axboe * must not do the output in nonblocking mode as then we 971b92ce558SJens Axboe * could get stuck data in the internal pipe: 972b92ce558SJens Axboe */ 973c66ab6faSJens Axboe ret = actor(pipe, sd); 974a82c53a0STom Zanussi if (unlikely(ret <= 0)) { 975a82c53a0STom Zanussi sd->pos = prev_pos; 976b92ce558SJens Axboe goto out_release; 977a82c53a0STom Zanussi } 978b92ce558SJens Axboe 979b92ce558SJens Axboe bytes += ret; 980b92ce558SJens Axboe len -= ret; 981bcd4f3acSJens Axboe sd->pos = pos; 982b92ce558SJens Axboe 983a82c53a0STom Zanussi if (ret < read_len) { 984a82c53a0STom Zanussi sd->pos = prev_pos + ret; 98551a92c0fSJens Axboe goto out_release; 986b92ce558SJens Axboe } 987a82c53a0STom Zanussi } 988b92ce558SJens Axboe 9899e97198dSJens Axboe done: 990b92ce558SJens Axboe pipe->nrbufs = pipe->curbuf = 0; 9919e97198dSJens Axboe file_accessed(in); 992b92ce558SJens Axboe return bytes; 993b92ce558SJens Axboe 994b92ce558SJens Axboe out_release: 995b92ce558SJens Axboe /* 996b92ce558SJens Axboe * If we did an incomplete transfer we must release 997b92ce558SJens Axboe * the pipe buffers in question: 998b92ce558SJens Axboe */ 99935f3d14dSJens Axboe for (i = 0; i < pipe->buffers; i++) { 1000b92ce558SJens Axboe struct pipe_buffer *buf = pipe->bufs + i; 1001b92ce558SJens Axboe 1002a779638cSMiklos Szeredi if (buf->ops) 1003a779638cSMiklos Szeredi pipe_buf_release(pipe, buf); 1004b92ce558SJens Axboe } 1005b92ce558SJens Axboe 10069e97198dSJens Axboe if (!bytes) 10079e97198dSJens Axboe bytes = ret; 1008b92ce558SJens Axboe 10099e97198dSJens Axboe goto done; 1010c66ab6faSJens Axboe } 1011c66ab6faSJens Axboe EXPORT_SYMBOL(splice_direct_to_actor); 1012c66ab6faSJens Axboe 1013c66ab6faSJens Axboe static int direct_splice_actor(struct pipe_inode_info *pipe, 1014c66ab6faSJens Axboe struct splice_desc *sd) 1015c66ab6faSJens Axboe { 10166a14b90bSJens Axboe struct file *file = sd->u.file; 1017c66ab6faSJens Axboe 10187995bd28SAl Viro return do_splice_from(pipe, file, sd->opos, sd->total_len, 10192cb4b05eSChangli Gao sd->flags); 1020c66ab6faSJens Axboe } 1021c66ab6faSJens Axboe 1022932cc6d4SJens Axboe /** 1023932cc6d4SJens Axboe * do_splice_direct - splices data directly between two files 1024932cc6d4SJens Axboe * @in: file to splice from 1025932cc6d4SJens Axboe * @ppos: input file offset 1026932cc6d4SJens Axboe * @out: file to splice to 1027acdb37c3SRandy Dunlap * @opos: output file offset 1028932cc6d4SJens Axboe * @len: number of bytes to splice 1029932cc6d4SJens Axboe * @flags: splice modifier flags 1030932cc6d4SJens Axboe * 1031932cc6d4SJens Axboe * Description: 1032932cc6d4SJens Axboe * For use by do_sendfile(). splice can easily emulate sendfile, but 1033932cc6d4SJens Axboe * doing it in the application would incur an extra system call 1034932cc6d4SJens Axboe * (splice in + splice out, as compared to just sendfile()). So this helper 1035932cc6d4SJens Axboe * can splice directly through a process-private pipe. 1036932cc6d4SJens Axboe * 1037932cc6d4SJens Axboe */ 1038c66ab6faSJens Axboe long do_splice_direct(struct file *in, loff_t *ppos, struct file *out, 10397995bd28SAl Viro loff_t *opos, size_t len, unsigned int flags) 1040c66ab6faSJens Axboe { 1041c66ab6faSJens Axboe struct splice_desc sd = { 1042c66ab6faSJens Axboe .len = len, 1043c66ab6faSJens Axboe .total_len = len, 1044c66ab6faSJens Axboe .flags = flags, 1045c66ab6faSJens Axboe .pos = *ppos, 10466a14b90bSJens Axboe .u.file = out, 10477995bd28SAl Viro .opos = opos, 1048c66ab6faSJens Axboe }; 104951a92c0fSJens Axboe long ret; 1050c66ab6faSJens Axboe 105118c67cb9SAl Viro if (unlikely(!(out->f_mode & FMODE_WRITE))) 105218c67cb9SAl Viro return -EBADF; 105318c67cb9SAl Viro 105418c67cb9SAl Viro if (unlikely(out->f_flags & O_APPEND)) 105518c67cb9SAl Viro return -EINVAL; 105618c67cb9SAl Viro 105718c67cb9SAl Viro ret = rw_verify_area(WRITE, out, opos, len); 105818c67cb9SAl Viro if (unlikely(ret < 0)) 105918c67cb9SAl Viro return ret; 106018c67cb9SAl Viro 1061c66ab6faSJens Axboe ret = splice_direct_to_actor(in, &sd, direct_splice_actor); 106251a92c0fSJens Axboe if (ret > 0) 1063a82c53a0STom Zanussi *ppos = sd.pos; 106451a92c0fSJens Axboe 1065c66ab6faSJens Axboe return ret; 1066b92ce558SJens Axboe } 10671c118596SMiklos Szeredi EXPORT_SYMBOL(do_splice_direct); 1068b92ce558SJens Axboe 10698924feffSAl Viro static int wait_for_space(struct pipe_inode_info *pipe, unsigned flags) 10708924feffSAl Viro { 107152bce911SLinus Torvalds for (;;) { 107252bce911SLinus Torvalds if (unlikely(!pipe->readers)) { 107352bce911SLinus Torvalds send_sig(SIGPIPE, current, 0); 107452bce911SLinus Torvalds return -EPIPE; 107552bce911SLinus Torvalds } 107652bce911SLinus Torvalds if (pipe->nrbufs != pipe->buffers) 107752bce911SLinus Torvalds return 0; 10788924feffSAl Viro if (flags & SPLICE_F_NONBLOCK) 10798924feffSAl Viro return -EAGAIN; 10808924feffSAl Viro if (signal_pending(current)) 10818924feffSAl Viro return -ERESTARTSYS; 10828924feffSAl Viro pipe->waiting_writers++; 10838924feffSAl Viro pipe_wait(pipe); 10848924feffSAl Viro pipe->waiting_writers--; 10858924feffSAl Viro } 10868924feffSAl Viro } 10878924feffSAl Viro 10887c77f0b3SMiklos Szeredi static int splice_pipe_to_pipe(struct pipe_inode_info *ipipe, 10897c77f0b3SMiklos Szeredi struct pipe_inode_info *opipe, 10907c77f0b3SMiklos Szeredi size_t len, unsigned int flags); 1091ddac0d39SJens Axboe 1092ddac0d39SJens Axboe /* 109383f9135bSJens Axboe * Determine where to splice to/from. 109483f9135bSJens Axboe */ 1095529565dcSIngo Molnar static long do_splice(struct file *in, loff_t __user *off_in, 1096529565dcSIngo Molnar struct file *out, loff_t __user *off_out, 1097529565dcSIngo Molnar size_t len, unsigned int flags) 10985274f052SJens Axboe { 10997c77f0b3SMiklos Szeredi struct pipe_inode_info *ipipe; 11007c77f0b3SMiklos Szeredi struct pipe_inode_info *opipe; 11017995bd28SAl Viro loff_t offset; 1102a4514ebdSJens Axboe long ret; 11035274f052SJens Axboe 110471993e62SLinus Torvalds ipipe = get_pipe_info(in); 110571993e62SLinus Torvalds opipe = get_pipe_info(out); 11067c77f0b3SMiklos Szeredi 11077c77f0b3SMiklos Szeredi if (ipipe && opipe) { 11087c77f0b3SMiklos Szeredi if (off_in || off_out) 11097c77f0b3SMiklos Szeredi return -ESPIPE; 11107c77f0b3SMiklos Szeredi 11117c77f0b3SMiklos Szeredi if (!(in->f_mode & FMODE_READ)) 11127c77f0b3SMiklos Szeredi return -EBADF; 11137c77f0b3SMiklos Szeredi 11147c77f0b3SMiklos Szeredi if (!(out->f_mode & FMODE_WRITE)) 11157c77f0b3SMiklos Szeredi return -EBADF; 11167c77f0b3SMiklos Szeredi 11177c77f0b3SMiklos Szeredi /* Splicing to self would be fun, but... */ 11187c77f0b3SMiklos Szeredi if (ipipe == opipe) 11197c77f0b3SMiklos Szeredi return -EINVAL; 11207c77f0b3SMiklos Szeredi 11217c77f0b3SMiklos Szeredi return splice_pipe_to_pipe(ipipe, opipe, len, flags); 11227c77f0b3SMiklos Szeredi } 11237c77f0b3SMiklos Szeredi 11247c77f0b3SMiklos Szeredi if (ipipe) { 1125529565dcSIngo Molnar if (off_in) 1126529565dcSIngo Molnar return -ESPIPE; 1127b92ce558SJens Axboe if (off_out) { 112819c9a49bSChangli Gao if (!(out->f_mode & FMODE_PWRITE)) 1129b92ce558SJens Axboe return -EINVAL; 1130cbb7e577SJens Axboe if (copy_from_user(&offset, off_out, sizeof(loff_t))) 1131b92ce558SJens Axboe return -EFAULT; 11327995bd28SAl Viro } else { 11337995bd28SAl Viro offset = out->f_pos; 11347995bd28SAl Viro } 1135529565dcSIngo Molnar 113618c67cb9SAl Viro if (unlikely(!(out->f_mode & FMODE_WRITE))) 113718c67cb9SAl Viro return -EBADF; 113818c67cb9SAl Viro 113918c67cb9SAl Viro if (unlikely(out->f_flags & O_APPEND)) 114018c67cb9SAl Viro return -EINVAL; 114118c67cb9SAl Viro 114218c67cb9SAl Viro ret = rw_verify_area(WRITE, out, &offset, len); 114318c67cb9SAl Viro if (unlikely(ret < 0)) 114418c67cb9SAl Viro return ret; 114518c67cb9SAl Viro 1146500368f7SAl Viro file_start_write(out); 11477995bd28SAl Viro ret = do_splice_from(ipipe, out, &offset, len, flags); 1148500368f7SAl Viro file_end_write(out); 1149a4514ebdSJens Axboe 11507995bd28SAl Viro if (!off_out) 11517995bd28SAl Viro out->f_pos = offset; 11527995bd28SAl Viro else if (copy_to_user(off_out, &offset, sizeof(loff_t))) 1153a4514ebdSJens Axboe ret = -EFAULT; 1154a4514ebdSJens Axboe 1155a4514ebdSJens Axboe return ret; 1156529565dcSIngo Molnar } 11575274f052SJens Axboe 11587c77f0b3SMiklos Szeredi if (opipe) { 1159529565dcSIngo Molnar if (off_out) 1160529565dcSIngo Molnar return -ESPIPE; 1161b92ce558SJens Axboe if (off_in) { 116219c9a49bSChangli Gao if (!(in->f_mode & FMODE_PREAD)) 1163b92ce558SJens Axboe return -EINVAL; 1164cbb7e577SJens Axboe if (copy_from_user(&offset, off_in, sizeof(loff_t))) 1165b92ce558SJens Axboe return -EFAULT; 11667995bd28SAl Viro } else { 11677995bd28SAl Viro offset = in->f_pos; 11687995bd28SAl Viro } 1169529565dcSIngo Molnar 11708924feffSAl Viro pipe_lock(opipe); 11718924feffSAl Viro ret = wait_for_space(opipe, flags); 11728924feffSAl Viro if (!ret) 11737995bd28SAl Viro ret = do_splice_to(in, &offset, opipe, len, flags); 11748924feffSAl Viro pipe_unlock(opipe); 11758924feffSAl Viro if (ret > 0) 11768924feffSAl Viro wakeup_pipe_readers(opipe); 11777995bd28SAl Viro if (!off_in) 11787995bd28SAl Viro in->f_pos = offset; 11797995bd28SAl Viro else if (copy_to_user(off_in, &offset, sizeof(loff_t))) 1180a4514ebdSJens Axboe ret = -EFAULT; 1181a4514ebdSJens Axboe 1182a4514ebdSJens Axboe return ret; 1183529565dcSIngo Molnar } 11845274f052SJens Axboe 11855274f052SJens Axboe return -EINVAL; 11865274f052SJens Axboe } 11875274f052SJens Axboe 118879fddc4eSAl Viro static int iter_to_pipe(struct iov_iter *from, 118979fddc4eSAl Viro struct pipe_inode_info *pipe, 119079fddc4eSAl Viro unsigned flags) 1191912d35f8SJens Axboe { 119279fddc4eSAl Viro struct pipe_buffer buf = { 119379fddc4eSAl Viro .ops = &user_page_pipe_buf_ops, 119479fddc4eSAl Viro .flags = flags 119579fddc4eSAl Viro }; 119679fddc4eSAl Viro size_t total = 0; 119779fddc4eSAl Viro int ret = 0; 119879fddc4eSAl Viro bool failed = false; 119979fddc4eSAl Viro 120079fddc4eSAl Viro while (iov_iter_count(from) && !failed) { 120179fddc4eSAl Viro struct page *pages[16]; 1202db85a9ebSAl Viro ssize_t copied; 1203db85a9ebSAl Viro size_t start; 120479fddc4eSAl Viro int n; 1205912d35f8SJens Axboe 120679fddc4eSAl Viro copied = iov_iter_get_pages(from, pages, ~0UL, 16, &start); 120779fddc4eSAl Viro if (copied <= 0) { 120879fddc4eSAl Viro ret = copied; 120979fddc4eSAl Viro break; 121079fddc4eSAl Viro } 1211912d35f8SJens Axboe 121279fddc4eSAl Viro for (n = 0; copied; n++, start = 0) { 1213db85a9ebSAl Viro int size = min_t(int, copied, PAGE_SIZE - start); 121479fddc4eSAl Viro if (!failed) { 121579fddc4eSAl Viro buf.page = pages[n]; 121679fddc4eSAl Viro buf.offset = start; 121779fddc4eSAl Viro buf.len = size; 121879fddc4eSAl Viro ret = add_to_pipe(pipe, &buf); 121979fddc4eSAl Viro if (unlikely(ret < 0)) { 122079fddc4eSAl Viro failed = true; 122179fddc4eSAl Viro } else { 122279fddc4eSAl Viro iov_iter_advance(from, ret); 122379fddc4eSAl Viro total += ret; 122479fddc4eSAl Viro } 122579fddc4eSAl Viro } else { 122679fddc4eSAl Viro put_page(pages[n]); 122779fddc4eSAl Viro } 1228db85a9ebSAl Viro copied -= size; 1229912d35f8SJens Axboe } 1230912d35f8SJens Axboe } 123179fddc4eSAl Viro return total ? total : ret; 1232912d35f8SJens Axboe } 1233912d35f8SJens Axboe 12346a14b90bSJens Axboe static int pipe_to_user(struct pipe_inode_info *pipe, struct pipe_buffer *buf, 12356a14b90bSJens Axboe struct splice_desc *sd) 12366a14b90bSJens Axboe { 12376130f531SAl Viro int n = copy_page_to_iter(buf->page, buf->offset, sd->len, sd->u.data); 12386130f531SAl Viro return n == sd->len ? n : -EFAULT; 12396a14b90bSJens Axboe } 12406a14b90bSJens Axboe 12416a14b90bSJens Axboe /* 12426a14b90bSJens Axboe * For lack of a better implementation, implement vmsplice() to userspace 12436a14b90bSJens Axboe * as a simple copy of the pipes pages to the user iov. 12446a14b90bSJens Axboe */ 124587a3002aSAl Viro static long vmsplice_to_user(struct file *file, struct iov_iter *iter, 124687a3002aSAl Viro unsigned int flags) 12476a14b90bSJens Axboe { 124887a3002aSAl Viro struct pipe_inode_info *pipe = get_pipe_info(file); 124987a3002aSAl Viro struct splice_desc sd = { 125087a3002aSAl Viro .total_len = iov_iter_count(iter), 125187a3002aSAl Viro .flags = flags, 125287a3002aSAl Viro .u.data = iter 125387a3002aSAl Viro }; 125487a3002aSAl Viro long ret = 0; 12556a14b90bSJens Axboe 12566a14b90bSJens Axboe if (!pipe) 12576a14b90bSJens Axboe return -EBADF; 12586a14b90bSJens Axboe 1259345995faSAl Viro if (sd.total_len) { 12606130f531SAl Viro pipe_lock(pipe); 12616130f531SAl Viro ret = __splice_from_pipe(pipe, &sd, pipe_to_user); 126261e0d47cSMiklos Szeredi pipe_unlock(pipe); 1263345995faSAl Viro } 12646a14b90bSJens Axboe 12656a14b90bSJens Axboe return ret; 12666a14b90bSJens Axboe } 12676a14b90bSJens Axboe 1268912d35f8SJens Axboe /* 1269912d35f8SJens Axboe * vmsplice splices a user address range into a pipe. It can be thought of 1270912d35f8SJens Axboe * as splice-from-memory, where the regular splice is splice-from-file (or 1271912d35f8SJens Axboe * to file). In both cases the output is a pipe, naturally. 1272912d35f8SJens Axboe */ 127387a3002aSAl Viro static long vmsplice_to_pipe(struct file *file, struct iov_iter *iter, 127487a3002aSAl Viro unsigned int flags) 1275912d35f8SJens Axboe { 1276ddac0d39SJens Axboe struct pipe_inode_info *pipe; 127787a3002aSAl Viro long ret = 0; 127879fddc4eSAl Viro unsigned buf_flag = 0; 127979fddc4eSAl Viro 128079fddc4eSAl Viro if (flags & SPLICE_F_GIFT) 128179fddc4eSAl Viro buf_flag = PIPE_BUF_FLAG_GIFT; 1282912d35f8SJens Axboe 128371993e62SLinus Torvalds pipe = get_pipe_info(file); 1284ddac0d39SJens Axboe if (!pipe) 1285912d35f8SJens Axboe return -EBADF; 1286912d35f8SJens Axboe 12878924feffSAl Viro pipe_lock(pipe); 12888924feffSAl Viro ret = wait_for_space(pipe, flags); 128979fddc4eSAl Viro if (!ret) 129087a3002aSAl Viro ret = iter_to_pipe(iter, pipe, buf_flag); 12918924feffSAl Viro pipe_unlock(pipe); 12928924feffSAl Viro if (ret > 0) 12938924feffSAl Viro wakeup_pipe_readers(pipe); 129435f3d14dSJens Axboe return ret; 1295912d35f8SJens Axboe } 1296912d35f8SJens Axboe 129787a3002aSAl Viro static int vmsplice_type(struct fd f, int *type) 129887a3002aSAl Viro { 129987a3002aSAl Viro if (!f.file) 130087a3002aSAl Viro return -EBADF; 130187a3002aSAl Viro if (f.file->f_mode & FMODE_WRITE) { 130287a3002aSAl Viro *type = WRITE; 130387a3002aSAl Viro } else if (f.file->f_mode & FMODE_READ) { 130487a3002aSAl Viro *type = READ; 130587a3002aSAl Viro } else { 130687a3002aSAl Viro fdput(f); 130787a3002aSAl Viro return -EBADF; 130887a3002aSAl Viro } 130987a3002aSAl Viro return 0; 131087a3002aSAl Viro } 131187a3002aSAl Viro 13126a14b90bSJens Axboe /* 13136a14b90bSJens Axboe * Note that vmsplice only really supports true splicing _from_ user memory 13146a14b90bSJens Axboe * to a pipe, not the other way around. Splicing from user memory is a simple 13156a14b90bSJens Axboe * operation that can be supported without any funky alignment restrictions 13166a14b90bSJens Axboe * or nasty vm tricks. We simply map in the user memory and fill them into 13176a14b90bSJens Axboe * a pipe. The reverse isn't quite as easy, though. There are two possible 13186a14b90bSJens Axboe * solutions for that: 13196a14b90bSJens Axboe * 13206a14b90bSJens Axboe * - memcpy() the data internally, at which point we might as well just 13216a14b90bSJens Axboe * do a regular read() on the buffer anyway. 13226a14b90bSJens Axboe * - Lots of nasty vm tricks, that are neither fast nor flexible (it 13236a14b90bSJens Axboe * has restriction limitations on both ends of the pipe). 13246a14b90bSJens Axboe * 13256a14b90bSJens Axboe * Currently we punt and implement it as a normal copy, see pipe_to_user(). 13266a14b90bSJens Axboe * 13276a14b90bSJens Axboe */ 132887a3002aSAl Viro static long do_vmsplice(struct file *f, struct iov_iter *iter, unsigned int flags) 1329912d35f8SJens Axboe { 13303d6ea290SAl Viro if (unlikely(flags & ~SPLICE_F_ALL)) 13313d6ea290SAl Viro return -EINVAL; 133287a3002aSAl Viro 133387a3002aSAl Viro if (!iov_iter_count(iter)) 13346a14b90bSJens Axboe return 0; 13356a14b90bSJens Axboe 133687a3002aSAl Viro if (iov_iter_rw(iter) == WRITE) 133787a3002aSAl Viro return vmsplice_to_pipe(f, iter, flags); 133887a3002aSAl Viro else 133987a3002aSAl Viro return vmsplice_to_user(f, iter, flags); 1340912d35f8SJens Axboe } 1341912d35f8SJens Axboe 134287a3002aSAl Viro SYSCALL_DEFINE4(vmsplice, int, fd, const struct iovec __user *, uiov, 134330cfe4efSDominik Brodowski unsigned long, nr_segs, unsigned int, flags) 134430cfe4efSDominik Brodowski { 134587a3002aSAl Viro struct iovec iovstack[UIO_FASTIOV]; 134687a3002aSAl Viro struct iovec *iov = iovstack; 134787a3002aSAl Viro struct iov_iter iter; 134887a3002aSAl Viro long error; 134987a3002aSAl Viro struct fd f; 135087a3002aSAl Viro int type; 135187a3002aSAl Viro 135287a3002aSAl Viro f = fdget(fd); 135387a3002aSAl Viro error = vmsplice_type(f, &type); 135487a3002aSAl Viro if (error) 135587a3002aSAl Viro return error; 135687a3002aSAl Viro 135787a3002aSAl Viro error = import_iovec(type, uiov, nr_segs, 135887a3002aSAl Viro ARRAY_SIZE(iovstack), &iov, &iter); 135987a3002aSAl Viro if (!error) { 136087a3002aSAl Viro error = do_vmsplice(f.file, &iter, flags); 136187a3002aSAl Viro kfree(iov); 136287a3002aSAl Viro } 136387a3002aSAl Viro fdput(f); 136487a3002aSAl Viro return error; 136530cfe4efSDominik Brodowski } 136630cfe4efSDominik Brodowski 136776b021d0SAl Viro #ifdef CONFIG_COMPAT 136876b021d0SAl Viro COMPAT_SYSCALL_DEFINE4(vmsplice, int, fd, const struct compat_iovec __user *, iov32, 136976b021d0SAl Viro unsigned int, nr_segs, unsigned int, flags) 137076b021d0SAl Viro { 137187a3002aSAl Viro struct iovec iovstack[UIO_FASTIOV]; 137287a3002aSAl Viro struct iovec *iov = iovstack; 137387a3002aSAl Viro struct iov_iter iter; 137487a3002aSAl Viro long error; 137587a3002aSAl Viro struct fd f; 137687a3002aSAl Viro int type; 137787a3002aSAl Viro 137887a3002aSAl Viro f = fdget(fd); 137987a3002aSAl Viro error = vmsplice_type(f, &type); 138087a3002aSAl Viro if (error) 138187a3002aSAl Viro return error; 138287a3002aSAl Viro 138387a3002aSAl Viro error = compat_import_iovec(type, iov32, nr_segs, 138487a3002aSAl Viro ARRAY_SIZE(iovstack), &iov, &iter); 138587a3002aSAl Viro if (!error) { 138687a3002aSAl Viro error = do_vmsplice(f.file, &iter, flags); 138787a3002aSAl Viro kfree(iov); 138876b021d0SAl Viro } 138987a3002aSAl Viro fdput(f); 139087a3002aSAl Viro return error; 139176b021d0SAl Viro } 139276b021d0SAl Viro #endif 139376b021d0SAl Viro 1394836f92adSHeiko Carstens SYSCALL_DEFINE6(splice, int, fd_in, loff_t __user *, off_in, 1395836f92adSHeiko Carstens int, fd_out, loff_t __user *, off_out, 1396836f92adSHeiko Carstens size_t, len, unsigned int, flags) 13975274f052SJens Axboe { 13982903ff01SAl Viro struct fd in, out; 13995274f052SJens Axboe long error; 14005274f052SJens Axboe 14015274f052SJens Axboe if (unlikely(!len)) 14025274f052SJens Axboe return 0; 14035274f052SJens Axboe 14043d6ea290SAl Viro if (unlikely(flags & ~SPLICE_F_ALL)) 14053d6ea290SAl Viro return -EINVAL; 14063d6ea290SAl Viro 14075274f052SJens Axboe error = -EBADF; 14082903ff01SAl Viro in = fdget(fd_in); 14092903ff01SAl Viro if (in.file) { 14102903ff01SAl Viro if (in.file->f_mode & FMODE_READ) { 14112903ff01SAl Viro out = fdget(fd_out); 14122903ff01SAl Viro if (out.file) { 14132903ff01SAl Viro if (out.file->f_mode & FMODE_WRITE) 14142903ff01SAl Viro error = do_splice(in.file, off_in, 14152903ff01SAl Viro out.file, off_out, 1416529565dcSIngo Molnar len, flags); 14172903ff01SAl Viro fdput(out); 14185274f052SJens Axboe } 14195274f052SJens Axboe } 14202903ff01SAl Viro fdput(in); 14215274f052SJens Axboe } 14225274f052SJens Axboe return error; 14235274f052SJens Axboe } 142470524490SJens Axboe 142570524490SJens Axboe /* 1426aadd06e5SJens Axboe * Make sure there's data to read. Wait for input if we can, otherwise 1427aadd06e5SJens Axboe * return an appropriate error. 1428aadd06e5SJens Axboe */ 14297c77f0b3SMiklos Szeredi static int ipipe_prep(struct pipe_inode_info *pipe, unsigned int flags) 1430aadd06e5SJens Axboe { 1431aadd06e5SJens Axboe int ret; 1432aadd06e5SJens Axboe 1433aadd06e5SJens Axboe /* 1434aadd06e5SJens Axboe * Check ->nrbufs without the inode lock first. This function 1435aadd06e5SJens Axboe * is speculative anyways, so missing one is ok. 1436aadd06e5SJens Axboe */ 1437aadd06e5SJens Axboe if (pipe->nrbufs) 1438aadd06e5SJens Axboe return 0; 1439aadd06e5SJens Axboe 1440aadd06e5SJens Axboe ret = 0; 144161e0d47cSMiklos Szeredi pipe_lock(pipe); 1442aadd06e5SJens Axboe 1443aadd06e5SJens Axboe while (!pipe->nrbufs) { 1444aadd06e5SJens Axboe if (signal_pending(current)) { 1445aadd06e5SJens Axboe ret = -ERESTARTSYS; 1446aadd06e5SJens Axboe break; 1447aadd06e5SJens Axboe } 1448aadd06e5SJens Axboe if (!pipe->writers) 1449aadd06e5SJens Axboe break; 1450aadd06e5SJens Axboe if (!pipe->waiting_writers) { 1451aadd06e5SJens Axboe if (flags & SPLICE_F_NONBLOCK) { 1452aadd06e5SJens Axboe ret = -EAGAIN; 1453aadd06e5SJens Axboe break; 1454aadd06e5SJens Axboe } 1455aadd06e5SJens Axboe } 1456aadd06e5SJens Axboe pipe_wait(pipe); 1457aadd06e5SJens Axboe } 1458aadd06e5SJens Axboe 145961e0d47cSMiklos Szeredi pipe_unlock(pipe); 1460aadd06e5SJens Axboe return ret; 1461aadd06e5SJens Axboe } 1462aadd06e5SJens Axboe 1463aadd06e5SJens Axboe /* 1464aadd06e5SJens Axboe * Make sure there's writeable room. Wait for room if we can, otherwise 1465aadd06e5SJens Axboe * return an appropriate error. 1466aadd06e5SJens Axboe */ 14677c77f0b3SMiklos Szeredi static int opipe_prep(struct pipe_inode_info *pipe, unsigned int flags) 1468aadd06e5SJens Axboe { 1469aadd06e5SJens Axboe int ret; 1470aadd06e5SJens Axboe 1471aadd06e5SJens Axboe /* 1472aadd06e5SJens Axboe * Check ->nrbufs without the inode lock first. This function 1473aadd06e5SJens Axboe * is speculative anyways, so missing one is ok. 1474aadd06e5SJens Axboe */ 147535f3d14dSJens Axboe if (pipe->nrbufs < pipe->buffers) 1476aadd06e5SJens Axboe return 0; 1477aadd06e5SJens Axboe 1478aadd06e5SJens Axboe ret = 0; 147961e0d47cSMiklos Szeredi pipe_lock(pipe); 1480aadd06e5SJens Axboe 148135f3d14dSJens Axboe while (pipe->nrbufs >= pipe->buffers) { 1482aadd06e5SJens Axboe if (!pipe->readers) { 1483aadd06e5SJens Axboe send_sig(SIGPIPE, current, 0); 1484aadd06e5SJens Axboe ret = -EPIPE; 1485aadd06e5SJens Axboe break; 1486aadd06e5SJens Axboe } 1487aadd06e5SJens Axboe if (flags & SPLICE_F_NONBLOCK) { 1488aadd06e5SJens Axboe ret = -EAGAIN; 1489aadd06e5SJens Axboe break; 1490aadd06e5SJens Axboe } 1491aadd06e5SJens Axboe if (signal_pending(current)) { 1492aadd06e5SJens Axboe ret = -ERESTARTSYS; 1493aadd06e5SJens Axboe break; 1494aadd06e5SJens Axboe } 1495aadd06e5SJens Axboe pipe->waiting_writers++; 1496aadd06e5SJens Axboe pipe_wait(pipe); 1497aadd06e5SJens Axboe pipe->waiting_writers--; 1498aadd06e5SJens Axboe } 1499aadd06e5SJens Axboe 150061e0d47cSMiklos Szeredi pipe_unlock(pipe); 1501aadd06e5SJens Axboe return ret; 1502aadd06e5SJens Axboe } 1503aadd06e5SJens Axboe 1504aadd06e5SJens Axboe /* 15057c77f0b3SMiklos Szeredi * Splice contents of ipipe to opipe. 15067c77f0b3SMiklos Szeredi */ 15077c77f0b3SMiklos Szeredi static int splice_pipe_to_pipe(struct pipe_inode_info *ipipe, 15087c77f0b3SMiklos Szeredi struct pipe_inode_info *opipe, 15097c77f0b3SMiklos Szeredi size_t len, unsigned int flags) 15107c77f0b3SMiklos Szeredi { 15117c77f0b3SMiklos Szeredi struct pipe_buffer *ibuf, *obuf; 15127c77f0b3SMiklos Szeredi int ret = 0, nbuf; 15137c77f0b3SMiklos Szeredi bool input_wakeup = false; 15147c77f0b3SMiklos Szeredi 15157c77f0b3SMiklos Szeredi 15167c77f0b3SMiklos Szeredi retry: 15177c77f0b3SMiklos Szeredi ret = ipipe_prep(ipipe, flags); 15187c77f0b3SMiklos Szeredi if (ret) 15197c77f0b3SMiklos Szeredi return ret; 15207c77f0b3SMiklos Szeredi 15217c77f0b3SMiklos Szeredi ret = opipe_prep(opipe, flags); 15227c77f0b3SMiklos Szeredi if (ret) 15237c77f0b3SMiklos Szeredi return ret; 15247c77f0b3SMiklos Szeredi 15257c77f0b3SMiklos Szeredi /* 15267c77f0b3SMiklos Szeredi * Potential ABBA deadlock, work around it by ordering lock 15277c77f0b3SMiklos Szeredi * grabbing by pipe info address. Otherwise two different processes 15287c77f0b3SMiklos Szeredi * could deadlock (one doing tee from A -> B, the other from B -> A). 15297c77f0b3SMiklos Szeredi */ 15307c77f0b3SMiklos Szeredi pipe_double_lock(ipipe, opipe); 15317c77f0b3SMiklos Szeredi 15327c77f0b3SMiklos Szeredi do { 15337c77f0b3SMiklos Szeredi if (!opipe->readers) { 15347c77f0b3SMiklos Szeredi send_sig(SIGPIPE, current, 0); 15357c77f0b3SMiklos Szeredi if (!ret) 15367c77f0b3SMiklos Szeredi ret = -EPIPE; 15377c77f0b3SMiklos Szeredi break; 15387c77f0b3SMiklos Szeredi } 15397c77f0b3SMiklos Szeredi 15407c77f0b3SMiklos Szeredi if (!ipipe->nrbufs && !ipipe->writers) 15417c77f0b3SMiklos Szeredi break; 15427c77f0b3SMiklos Szeredi 15437c77f0b3SMiklos Szeredi /* 15447c77f0b3SMiklos Szeredi * Cannot make any progress, because either the input 15457c77f0b3SMiklos Szeredi * pipe is empty or the output pipe is full. 15467c77f0b3SMiklos Szeredi */ 154735f3d14dSJens Axboe if (!ipipe->nrbufs || opipe->nrbufs >= opipe->buffers) { 15487c77f0b3SMiklos Szeredi /* Already processed some buffers, break */ 15497c77f0b3SMiklos Szeredi if (ret) 15507c77f0b3SMiklos Szeredi break; 15517c77f0b3SMiklos Szeredi 15527c77f0b3SMiklos Szeredi if (flags & SPLICE_F_NONBLOCK) { 15537c77f0b3SMiklos Szeredi ret = -EAGAIN; 15547c77f0b3SMiklos Szeredi break; 15557c77f0b3SMiklos Szeredi } 15567c77f0b3SMiklos Szeredi 15577c77f0b3SMiklos Szeredi /* 15587c77f0b3SMiklos Szeredi * We raced with another reader/writer and haven't 15597c77f0b3SMiklos Szeredi * managed to process any buffers. A zero return 15607c77f0b3SMiklos Szeredi * value means EOF, so retry instead. 15617c77f0b3SMiklos Szeredi */ 15627c77f0b3SMiklos Szeredi pipe_unlock(ipipe); 15637c77f0b3SMiklos Szeredi pipe_unlock(opipe); 15647c77f0b3SMiklos Szeredi goto retry; 15657c77f0b3SMiklos Szeredi } 15667c77f0b3SMiklos Szeredi 15677c77f0b3SMiklos Szeredi ibuf = ipipe->bufs + ipipe->curbuf; 156835f3d14dSJens Axboe nbuf = (opipe->curbuf + opipe->nrbufs) & (opipe->buffers - 1); 15697c77f0b3SMiklos Szeredi obuf = opipe->bufs + nbuf; 15707c77f0b3SMiklos Szeredi 15717c77f0b3SMiklos Szeredi if (len >= ibuf->len) { 15727c77f0b3SMiklos Szeredi /* 15737c77f0b3SMiklos Szeredi * Simply move the whole buffer from ipipe to opipe 15747c77f0b3SMiklos Szeredi */ 15757c77f0b3SMiklos Szeredi *obuf = *ibuf; 15767c77f0b3SMiklos Szeredi ibuf->ops = NULL; 15777c77f0b3SMiklos Szeredi opipe->nrbufs++; 157835f3d14dSJens Axboe ipipe->curbuf = (ipipe->curbuf + 1) & (ipipe->buffers - 1); 15797c77f0b3SMiklos Szeredi ipipe->nrbufs--; 15807c77f0b3SMiklos Szeredi input_wakeup = true; 15817c77f0b3SMiklos Szeredi } else { 15827c77f0b3SMiklos Szeredi /* 15837c77f0b3SMiklos Szeredi * Get a reference to this pipe buffer, 15847c77f0b3SMiklos Szeredi * so we can copy the contents over. 15857c77f0b3SMiklos Szeredi */ 15867bf2d1dfSMiklos Szeredi pipe_buf_get(ipipe, ibuf); 15877c77f0b3SMiklos Szeredi *obuf = *ibuf; 15887c77f0b3SMiklos Szeredi 15897c77f0b3SMiklos Szeredi /* 15907c77f0b3SMiklos Szeredi * Don't inherit the gift flag, we need to 15917c77f0b3SMiklos Szeredi * prevent multiple steals of this page. 15927c77f0b3SMiklos Szeredi */ 15937c77f0b3SMiklos Szeredi obuf->flags &= ~PIPE_BUF_FLAG_GIFT; 15947c77f0b3SMiklos Szeredi 15957c77f0b3SMiklos Szeredi obuf->len = len; 15967c77f0b3SMiklos Szeredi opipe->nrbufs++; 15977c77f0b3SMiklos Szeredi ibuf->offset += obuf->len; 15987c77f0b3SMiklos Szeredi ibuf->len -= obuf->len; 15997c77f0b3SMiklos Szeredi } 16007c77f0b3SMiklos Szeredi ret += obuf->len; 16017c77f0b3SMiklos Szeredi len -= obuf->len; 16027c77f0b3SMiklos Szeredi } while (len); 16037c77f0b3SMiklos Szeredi 16047c77f0b3SMiklos Szeredi pipe_unlock(ipipe); 16057c77f0b3SMiklos Szeredi pipe_unlock(opipe); 16067c77f0b3SMiklos Szeredi 16077c77f0b3SMiklos Szeredi /* 16087c77f0b3SMiklos Szeredi * If we put data in the output pipe, wakeup any potential readers. 16097c77f0b3SMiklos Szeredi */ 1610825cdcb1SNamhyung Kim if (ret > 0) 1611825cdcb1SNamhyung Kim wakeup_pipe_readers(opipe); 1612825cdcb1SNamhyung Kim 16137c77f0b3SMiklos Szeredi if (input_wakeup) 16147c77f0b3SMiklos Szeredi wakeup_pipe_writers(ipipe); 16157c77f0b3SMiklos Szeredi 16167c77f0b3SMiklos Szeredi return ret; 16177c77f0b3SMiklos Szeredi } 16187c77f0b3SMiklos Szeredi 16197c77f0b3SMiklos Szeredi /* 162070524490SJens Axboe * Link contents of ipipe to opipe. 162170524490SJens Axboe */ 162270524490SJens Axboe static int link_pipe(struct pipe_inode_info *ipipe, 162370524490SJens Axboe struct pipe_inode_info *opipe, 162470524490SJens Axboe size_t len, unsigned int flags) 162570524490SJens Axboe { 162670524490SJens Axboe struct pipe_buffer *ibuf, *obuf; 1627aadd06e5SJens Axboe int ret = 0, i = 0, nbuf; 162870524490SJens Axboe 162970524490SJens Axboe /* 163070524490SJens Axboe * Potential ABBA deadlock, work around it by ordering lock 163161e0d47cSMiklos Szeredi * grabbing by pipe info address. Otherwise two different processes 163270524490SJens Axboe * could deadlock (one doing tee from A -> B, the other from B -> A). 163370524490SJens Axboe */ 163461e0d47cSMiklos Szeredi pipe_double_lock(ipipe, opipe); 163570524490SJens Axboe 1636aadd06e5SJens Axboe do { 163770524490SJens Axboe if (!opipe->readers) { 163870524490SJens Axboe send_sig(SIGPIPE, current, 0); 163970524490SJens Axboe if (!ret) 164070524490SJens Axboe ret = -EPIPE; 164170524490SJens Axboe break; 164270524490SJens Axboe } 164370524490SJens Axboe 164470524490SJens Axboe /* 1645aadd06e5SJens Axboe * If we have iterated all input buffers or ran out of 1646aadd06e5SJens Axboe * output room, break. 164770524490SJens Axboe */ 164835f3d14dSJens Axboe if (i >= ipipe->nrbufs || opipe->nrbufs >= opipe->buffers) 1649aadd06e5SJens Axboe break; 1650aadd06e5SJens Axboe 165135f3d14dSJens Axboe ibuf = ipipe->bufs + ((ipipe->curbuf + i) & (ipipe->buffers-1)); 165235f3d14dSJens Axboe nbuf = (opipe->curbuf + opipe->nrbufs) & (opipe->buffers - 1); 165370524490SJens Axboe 165470524490SJens Axboe /* 165570524490SJens Axboe * Get a reference to this pipe buffer, 165670524490SJens Axboe * so we can copy the contents over. 165770524490SJens Axboe */ 16587bf2d1dfSMiklos Szeredi pipe_buf_get(ipipe, ibuf); 165970524490SJens Axboe 166070524490SJens Axboe obuf = opipe->bufs + nbuf; 166170524490SJens Axboe *obuf = *ibuf; 166270524490SJens Axboe 16637afa6fd0SJens Axboe /* 16647afa6fd0SJens Axboe * Don't inherit the gift flag, we need to 16657afa6fd0SJens Axboe * prevent multiple steals of this page. 16667afa6fd0SJens Axboe */ 16677afa6fd0SJens Axboe obuf->flags &= ~PIPE_BUF_FLAG_GIFT; 16687afa6fd0SJens Axboe 166970524490SJens Axboe if (obuf->len > len) 167070524490SJens Axboe obuf->len = len; 167170524490SJens Axboe 167270524490SJens Axboe opipe->nrbufs++; 167370524490SJens Axboe ret += obuf->len; 167470524490SJens Axboe len -= obuf->len; 1675aadd06e5SJens Axboe i++; 1676aadd06e5SJens Axboe } while (len); 167770524490SJens Axboe 167802cf01aeSJens Axboe /* 167902cf01aeSJens Axboe * return EAGAIN if we have the potential of some data in the 168002cf01aeSJens Axboe * future, otherwise just return 0 168102cf01aeSJens Axboe */ 168202cf01aeSJens Axboe if (!ret && ipipe->waiting_writers && (flags & SPLICE_F_NONBLOCK)) 168302cf01aeSJens Axboe ret = -EAGAIN; 168402cf01aeSJens Axboe 168561e0d47cSMiklos Szeredi pipe_unlock(ipipe); 168661e0d47cSMiklos Szeredi pipe_unlock(opipe); 168770524490SJens Axboe 1688aadd06e5SJens Axboe /* 1689aadd06e5SJens Axboe * If we put data in the output pipe, wakeup any potential readers. 1690aadd06e5SJens Axboe */ 1691825cdcb1SNamhyung Kim if (ret > 0) 1692825cdcb1SNamhyung Kim wakeup_pipe_readers(opipe); 169370524490SJens Axboe 169470524490SJens Axboe return ret; 169570524490SJens Axboe } 169670524490SJens Axboe 169770524490SJens Axboe /* 169870524490SJens Axboe * This is a tee(1) implementation that works on pipes. It doesn't copy 169970524490SJens Axboe * any data, it simply references the 'in' pages on the 'out' pipe. 170070524490SJens Axboe * The 'flags' used are the SPLICE_F_* variants, currently the only 170170524490SJens Axboe * applicable one is SPLICE_F_NONBLOCK. 170270524490SJens Axboe */ 170370524490SJens Axboe static long do_tee(struct file *in, struct file *out, size_t len, 170470524490SJens Axboe unsigned int flags) 170570524490SJens Axboe { 170671993e62SLinus Torvalds struct pipe_inode_info *ipipe = get_pipe_info(in); 170771993e62SLinus Torvalds struct pipe_inode_info *opipe = get_pipe_info(out); 1708aadd06e5SJens Axboe int ret = -EINVAL; 170970524490SJens Axboe 171070524490SJens Axboe /* 1711aadd06e5SJens Axboe * Duplicate the contents of ipipe to opipe without actually 1712aadd06e5SJens Axboe * copying the data. 171370524490SJens Axboe */ 1714aadd06e5SJens Axboe if (ipipe && opipe && ipipe != opipe) { 1715aadd06e5SJens Axboe /* 1716aadd06e5SJens Axboe * Keep going, unless we encounter an error. The ipipe/opipe 1717aadd06e5SJens Axboe * ordering doesn't really matter. 1718aadd06e5SJens Axboe */ 17197c77f0b3SMiklos Szeredi ret = ipipe_prep(ipipe, flags); 1720aadd06e5SJens Axboe if (!ret) { 17217c77f0b3SMiklos Szeredi ret = opipe_prep(opipe, flags); 172202cf01aeSJens Axboe if (!ret) 1723aadd06e5SJens Axboe ret = link_pipe(ipipe, opipe, len, flags); 1724aadd06e5SJens Axboe } 1725aadd06e5SJens Axboe } 172670524490SJens Axboe 1727aadd06e5SJens Axboe return ret; 172870524490SJens Axboe } 172970524490SJens Axboe 1730836f92adSHeiko Carstens SYSCALL_DEFINE4(tee, int, fdin, int, fdout, size_t, len, unsigned int, flags) 173170524490SJens Axboe { 17322903ff01SAl Viro struct fd in; 17332903ff01SAl Viro int error; 173470524490SJens Axboe 17353d6ea290SAl Viro if (unlikely(flags & ~SPLICE_F_ALL)) 17363d6ea290SAl Viro return -EINVAL; 17373d6ea290SAl Viro 173870524490SJens Axboe if (unlikely(!len)) 173970524490SJens Axboe return 0; 174070524490SJens Axboe 174170524490SJens Axboe error = -EBADF; 17422903ff01SAl Viro in = fdget(fdin); 17432903ff01SAl Viro if (in.file) { 17442903ff01SAl Viro if (in.file->f_mode & FMODE_READ) { 17452903ff01SAl Viro struct fd out = fdget(fdout); 17462903ff01SAl Viro if (out.file) { 17472903ff01SAl Viro if (out.file->f_mode & FMODE_WRITE) 17482903ff01SAl Viro error = do_tee(in.file, out.file, 17492903ff01SAl Viro len, flags); 17502903ff01SAl Viro fdput(out); 175170524490SJens Axboe } 175270524490SJens Axboe } 17532903ff01SAl Viro fdput(in); 175470524490SJens Axboe } 175570524490SJens Axboe 175670524490SJens Axboe return error; 175770524490SJens Axboe } 1758