Linux Audio

Check our new training course

Loading...
v4.17
   1// SPDX-License-Identifier: GPL-2.0
   2/*
   3 *  linux/fs/pipe.c
   4 *
   5 *  Copyright (C) 1991, 1992, 1999  Linus Torvalds
   6 */
   7
   8#include <linux/mm.h>
   9#include <linux/file.h>
  10#include <linux/poll.h>
  11#include <linux/slab.h>
  12#include <linux/module.h>
  13#include <linux/init.h>
  14#include <linux/fs.h>
  15#include <linux/log2.h>
  16#include <linux/mount.h>
  17#include <linux/magic.h>
  18#include <linux/pipe_fs_i.h>
  19#include <linux/uio.h>
  20#include <linux/highmem.h>
  21#include <linux/pagemap.h>
  22#include <linux/audit.h>
  23#include <linux/syscalls.h>
  24#include <linux/fcntl.h>
  25#include <linux/memcontrol.h>
  26
  27#include <linux/uaccess.h>
  28#include <asm/ioctls.h>
  29
  30#include "internal.h"
  31
  32/*
  33 * The max size that a non-root user is allowed to grow the pipe. Can
  34 * be set by root in /proc/sys/fs/pipe-max-size
  35 */
  36unsigned int pipe_max_size = 1048576;
  37
 
 
 
 
 
  38/* Maximum allocatable pages per user. Hard limit is unset by default, soft
  39 * matches default values.
  40 */
  41unsigned long pipe_user_pages_hard;
  42unsigned long pipe_user_pages_soft = PIPE_DEF_BUFFERS * INR_OPEN_CUR;
  43
  44/*
  45 * We use a start+len construction, which provides full use of the 
  46 * allocated memory.
  47 * -- Florian Coosmann (FGC)
  48 * 
  49 * Reads with count = 0 should always return 0.
  50 * -- Julian Bradfield 1999-06-07.
  51 *
  52 * FIFOs and Pipes now generate SIGIO for both readers and writers.
  53 * -- Jeremy Elson <jelson@circlemud.org> 2001-08-16
  54 *
  55 * pipe_read & write cleanup
  56 * -- Manfred Spraul <manfred@colorfullife.com> 2002-05-09
  57 */
  58
  59static void pipe_lock_nested(struct pipe_inode_info *pipe, int subclass)
  60{
  61	if (pipe->files)
  62		mutex_lock_nested(&pipe->mutex, subclass);
  63}
  64
  65void pipe_lock(struct pipe_inode_info *pipe)
  66{
  67	/*
  68	 * pipe_lock() nests non-pipe inode locks (for writing to a file)
  69	 */
  70	pipe_lock_nested(pipe, I_MUTEX_PARENT);
  71}
  72EXPORT_SYMBOL(pipe_lock);
  73
  74void pipe_unlock(struct pipe_inode_info *pipe)
  75{
  76	if (pipe->files)
  77		mutex_unlock(&pipe->mutex);
  78}
  79EXPORT_SYMBOL(pipe_unlock);
  80
  81static inline void __pipe_lock(struct pipe_inode_info *pipe)
  82{
  83	mutex_lock_nested(&pipe->mutex, I_MUTEX_PARENT);
  84}
  85
  86static inline void __pipe_unlock(struct pipe_inode_info *pipe)
  87{
  88	mutex_unlock(&pipe->mutex);
  89}
  90
  91void pipe_double_lock(struct pipe_inode_info *pipe1,
  92		      struct pipe_inode_info *pipe2)
  93{
  94	BUG_ON(pipe1 == pipe2);
  95
  96	if (pipe1 < pipe2) {
  97		pipe_lock_nested(pipe1, I_MUTEX_PARENT);
  98		pipe_lock_nested(pipe2, I_MUTEX_CHILD);
  99	} else {
 100		pipe_lock_nested(pipe2, I_MUTEX_PARENT);
 101		pipe_lock_nested(pipe1, I_MUTEX_CHILD);
 102	}
 103}
 104
 105/* Drop the inode semaphore and wait for a pipe event, atomically */
 106void pipe_wait(struct pipe_inode_info *pipe)
 107{
 108	DEFINE_WAIT(wait);
 109
 110	/*
 111	 * Pipes are system-local resources, so sleeping on them
 112	 * is considered a noninteractive wait:
 113	 */
 114	prepare_to_wait(&pipe->wait, &wait, TASK_INTERRUPTIBLE);
 115	pipe_unlock(pipe);
 116	schedule();
 117	finish_wait(&pipe->wait, &wait);
 118	pipe_lock(pipe);
 119}
 120
 121static void anon_pipe_buf_release(struct pipe_inode_info *pipe,
 122				  struct pipe_buffer *buf)
 123{
 124	struct page *page = buf->page;
 125
 126	/*
 127	 * If nobody else uses this page, and we don't already have a
 128	 * temporary page, let's keep track of it as a one-deep
 129	 * allocation cache. (Otherwise just release our reference to it)
 130	 */
 131	if (page_count(page) == 1 && !pipe->tmp_page)
 132		pipe->tmp_page = page;
 133	else
 134		put_page(page);
 135}
 136
 137static int anon_pipe_buf_steal(struct pipe_inode_info *pipe,
 138			       struct pipe_buffer *buf)
 139{
 140	struct page *page = buf->page;
 141
 142	if (page_count(page) == 1) {
 143		if (memcg_kmem_enabled())
 144			memcg_kmem_uncharge(page, 0);
 145		__SetPageLocked(page);
 146		return 0;
 147	}
 148	return 1;
 149}
 150
 151/**
 152 * generic_pipe_buf_steal - attempt to take ownership of a &pipe_buffer
 153 * @pipe:	the pipe that the buffer belongs to
 154 * @buf:	the buffer to attempt to steal
 155 *
 156 * Description:
 157 *	This function attempts to steal the &struct page attached to
 158 *	@buf. If successful, this function returns 0 and returns with
 159 *	the page locked. The caller may then reuse the page for whatever
 160 *	he wishes; the typical use is insertion into a different file
 161 *	page cache.
 162 */
 163int generic_pipe_buf_steal(struct pipe_inode_info *pipe,
 164			   struct pipe_buffer *buf)
 165{
 166	struct page *page = buf->page;
 167
 168	/*
 169	 * A reference of one is golden, that means that the owner of this
 170	 * page is the only one holding a reference to it. lock the page
 171	 * and return OK.
 172	 */
 173	if (page_count(page) == 1) {
 174		lock_page(page);
 175		return 0;
 176	}
 177
 178	return 1;
 179}
 180EXPORT_SYMBOL(generic_pipe_buf_steal);
 181
 182/**
 183 * generic_pipe_buf_get - get a reference to a &struct pipe_buffer
 184 * @pipe:	the pipe that the buffer belongs to
 185 * @buf:	the buffer to get a reference to
 186 *
 187 * Description:
 188 *	This function grabs an extra reference to @buf. It's used in
 189 *	in the tee() system call, when we duplicate the buffers in one
 190 *	pipe into another.
 191 */
 192void generic_pipe_buf_get(struct pipe_inode_info *pipe, struct pipe_buffer *buf)
 193{
 194	get_page(buf->page);
 195}
 196EXPORT_SYMBOL(generic_pipe_buf_get);
 197
 198/**
 199 * generic_pipe_buf_confirm - verify contents of the pipe buffer
 200 * @info:	the pipe that the buffer belongs to
 201 * @buf:	the buffer to confirm
 202 *
 203 * Description:
 204 *	This function does nothing, because the generic pipe code uses
 205 *	pages that are always good when inserted into the pipe.
 206 */
 207int generic_pipe_buf_confirm(struct pipe_inode_info *info,
 208			     struct pipe_buffer *buf)
 209{
 210	return 0;
 211}
 212EXPORT_SYMBOL(generic_pipe_buf_confirm);
 213
 214/**
 215 * generic_pipe_buf_release - put a reference to a &struct pipe_buffer
 216 * @pipe:	the pipe that the buffer belongs to
 217 * @buf:	the buffer to put a reference to
 218 *
 219 * Description:
 220 *	This function releases a reference to @buf.
 221 */
 222void generic_pipe_buf_release(struct pipe_inode_info *pipe,
 223			      struct pipe_buffer *buf)
 224{
 225	put_page(buf->page);
 226}
 227EXPORT_SYMBOL(generic_pipe_buf_release);
 228
 229static const struct pipe_buf_operations anon_pipe_buf_ops = {
 230	.can_merge = 1,
 231	.confirm = generic_pipe_buf_confirm,
 232	.release = anon_pipe_buf_release,
 233	.steal = anon_pipe_buf_steal,
 234	.get = generic_pipe_buf_get,
 235};
 236
 237static const struct pipe_buf_operations packet_pipe_buf_ops = {
 238	.can_merge = 0,
 239	.confirm = generic_pipe_buf_confirm,
 240	.release = anon_pipe_buf_release,
 241	.steal = anon_pipe_buf_steal,
 242	.get = generic_pipe_buf_get,
 243};
 244
 245static ssize_t
 246pipe_read(struct kiocb *iocb, struct iov_iter *to)
 247{
 248	size_t total_len = iov_iter_count(to);
 249	struct file *filp = iocb->ki_filp;
 250	struct pipe_inode_info *pipe = filp->private_data;
 251	int do_wakeup;
 252	ssize_t ret;
 253
 254	/* Null read succeeds. */
 255	if (unlikely(total_len == 0))
 256		return 0;
 257
 258	do_wakeup = 0;
 259	ret = 0;
 260	__pipe_lock(pipe);
 261	for (;;) {
 262		int bufs = pipe->nrbufs;
 263		if (bufs) {
 264			int curbuf = pipe->curbuf;
 265			struct pipe_buffer *buf = pipe->bufs + curbuf;
 
 266			size_t chars = buf->len;
 267			size_t written;
 268			int error;
 269
 270			if (chars > total_len)
 271				chars = total_len;
 272
 273			error = pipe_buf_confirm(pipe, buf);
 274			if (error) {
 275				if (!ret)
 276					ret = error;
 277				break;
 278			}
 279
 280			written = copy_page_to_iter(buf->page, buf->offset, chars, to);
 281			if (unlikely(written < chars)) {
 282				if (!ret)
 283					ret = -EFAULT;
 284				break;
 285			}
 286			ret += chars;
 287			buf->offset += chars;
 288			buf->len -= chars;
 289
 290			/* Was it a packet buffer? Clean up and exit */
 291			if (buf->flags & PIPE_BUF_FLAG_PACKET) {
 292				total_len = chars;
 293				buf->len = 0;
 294			}
 295
 296			if (!buf->len) {
 297				pipe_buf_release(pipe, buf);
 
 298				curbuf = (curbuf + 1) & (pipe->buffers - 1);
 299				pipe->curbuf = curbuf;
 300				pipe->nrbufs = --bufs;
 301				do_wakeup = 1;
 302			}
 303			total_len -= chars;
 304			if (!total_len)
 305				break;	/* common path: read succeeded */
 306		}
 307		if (bufs)	/* More to do? */
 308			continue;
 309		if (!pipe->writers)
 310			break;
 311		if (!pipe->waiting_writers) {
 312			/* syscall merging: Usually we must not sleep
 313			 * if O_NONBLOCK is set, or if we got some data.
 314			 * But if a writer sleeps in kernel space, then
 315			 * we can wait for that data without violating POSIX.
 316			 */
 317			if (ret)
 318				break;
 319			if (filp->f_flags & O_NONBLOCK) {
 320				ret = -EAGAIN;
 321				break;
 322			}
 323		}
 324		if (signal_pending(current)) {
 325			if (!ret)
 326				ret = -ERESTARTSYS;
 327			break;
 328		}
 329		if (do_wakeup) {
 330			wake_up_interruptible_sync_poll(&pipe->wait, EPOLLOUT | EPOLLWRNORM);
 331 			kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
 332		}
 333		pipe_wait(pipe);
 334	}
 335	__pipe_unlock(pipe);
 336
 337	/* Signal writers asynchronously that there is more room. */
 338	if (do_wakeup) {
 339		wake_up_interruptible_sync_poll(&pipe->wait, EPOLLOUT | EPOLLWRNORM);
 340		kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
 341	}
 342	if (ret > 0)
 343		file_accessed(filp);
 344	return ret;
 345}
 346
 347static inline int is_packetized(struct file *file)
 348{
 349	return (file->f_flags & O_DIRECT) != 0;
 350}
 351
 352static ssize_t
 353pipe_write(struct kiocb *iocb, struct iov_iter *from)
 354{
 355	struct file *filp = iocb->ki_filp;
 356	struct pipe_inode_info *pipe = filp->private_data;
 357	ssize_t ret = 0;
 358	int do_wakeup = 0;
 359	size_t total_len = iov_iter_count(from);
 360	ssize_t chars;
 361
 362	/* Null write succeeds. */
 363	if (unlikely(total_len == 0))
 364		return 0;
 365
 366	__pipe_lock(pipe);
 367
 368	if (!pipe->readers) {
 369		send_sig(SIGPIPE, current, 0);
 370		ret = -EPIPE;
 371		goto out;
 372	}
 373
 374	/* We try to merge small writes */
 375	chars = total_len & (PAGE_SIZE-1); /* size of the last buffer */
 376	if (pipe->nrbufs && chars != 0) {
 377		int lastbuf = (pipe->curbuf + pipe->nrbufs - 1) &
 378							(pipe->buffers - 1);
 379		struct pipe_buffer *buf = pipe->bufs + lastbuf;
 
 380		int offset = buf->offset + buf->len;
 381
 382		if (buf->ops->can_merge && offset + chars <= PAGE_SIZE) {
 383			ret = pipe_buf_confirm(pipe, buf);
 384			if (ret)
 385				goto out;
 386
 387			ret = copy_page_from_iter(buf->page, offset, chars, from);
 388			if (unlikely(ret < chars)) {
 389				ret = -EFAULT;
 390				goto out;
 391			}
 392			do_wakeup = 1;
 393			buf->len += ret;
 394			if (!iov_iter_count(from))
 395				goto out;
 396		}
 397	}
 398
 399	for (;;) {
 400		int bufs;
 401
 402		if (!pipe->readers) {
 403			send_sig(SIGPIPE, current, 0);
 404			if (!ret)
 405				ret = -EPIPE;
 406			break;
 407		}
 408		bufs = pipe->nrbufs;
 409		if (bufs < pipe->buffers) {
 410			int newbuf = (pipe->curbuf + bufs) & (pipe->buffers-1);
 411			struct pipe_buffer *buf = pipe->bufs + newbuf;
 412			struct page *page = pipe->tmp_page;
 413			int copied;
 414
 415			if (!page) {
 416				page = alloc_page(GFP_HIGHUSER | __GFP_ACCOUNT);
 417				if (unlikely(!page)) {
 418					ret = ret ? : -ENOMEM;
 419					break;
 420				}
 421				pipe->tmp_page = page;
 422			}
 423			/* Always wake up, even if the copy fails. Otherwise
 424			 * we lock up (O_NONBLOCK-)readers that sleep due to
 425			 * syscall merging.
 426			 * FIXME! Is this really true?
 427			 */
 428			do_wakeup = 1;
 429			copied = copy_page_from_iter(page, 0, PAGE_SIZE, from);
 430			if (unlikely(copied < PAGE_SIZE && iov_iter_count(from))) {
 431				if (!ret)
 432					ret = -EFAULT;
 433				break;
 434			}
 435			ret += copied;
 436
 437			/* Insert it into the buffer array */
 438			buf->page = page;
 439			buf->ops = &anon_pipe_buf_ops;
 440			buf->offset = 0;
 441			buf->len = copied;
 442			buf->flags = 0;
 443			if (is_packetized(filp)) {
 444				buf->ops = &packet_pipe_buf_ops;
 445				buf->flags = PIPE_BUF_FLAG_PACKET;
 446			}
 447			pipe->nrbufs = ++bufs;
 448			pipe->tmp_page = NULL;
 449
 450			if (!iov_iter_count(from))
 451				break;
 452		}
 453		if (bufs < pipe->buffers)
 454			continue;
 455		if (filp->f_flags & O_NONBLOCK) {
 456			if (!ret)
 457				ret = -EAGAIN;
 458			break;
 459		}
 460		if (signal_pending(current)) {
 461			if (!ret)
 462				ret = -ERESTARTSYS;
 463			break;
 464		}
 465		if (do_wakeup) {
 466			wake_up_interruptible_sync_poll(&pipe->wait, EPOLLIN | EPOLLRDNORM);
 467			kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
 468			do_wakeup = 0;
 469		}
 470		pipe->waiting_writers++;
 471		pipe_wait(pipe);
 472		pipe->waiting_writers--;
 473	}
 474out:
 475	__pipe_unlock(pipe);
 476	if (do_wakeup) {
 477		wake_up_interruptible_sync_poll(&pipe->wait, EPOLLIN | EPOLLRDNORM);
 478		kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
 479	}
 480	if (ret > 0 && sb_start_write_trylock(file_inode(filp)->i_sb)) {
 481		int err = file_update_time(filp);
 482		if (err)
 483			ret = err;
 484		sb_end_write(file_inode(filp)->i_sb);
 485	}
 486	return ret;
 487}
 488
 489static long pipe_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
 490{
 491	struct pipe_inode_info *pipe = filp->private_data;
 492	int count, buf, nrbufs;
 493
 494	switch (cmd) {
 495		case FIONREAD:
 496			__pipe_lock(pipe);
 497			count = 0;
 498			buf = pipe->curbuf;
 499			nrbufs = pipe->nrbufs;
 500			while (--nrbufs >= 0) {
 501				count += pipe->bufs[buf].len;
 502				buf = (buf+1) & (pipe->buffers - 1);
 503			}
 504			__pipe_unlock(pipe);
 505
 506			return put_user(count, (int __user *)arg);
 507		default:
 508			return -ENOIOCTLCMD;
 509	}
 510}
 511
 512/* No kernel lock held - fine */
 513static __poll_t
 514pipe_poll(struct file *filp, poll_table *wait)
 515{
 516	__poll_t mask;
 517	struct pipe_inode_info *pipe = filp->private_data;
 518	int nrbufs;
 519
 520	poll_wait(filp, &pipe->wait, wait);
 521
 522	/* Reading only -- no need for acquiring the semaphore.  */
 523	nrbufs = pipe->nrbufs;
 524	mask = 0;
 525	if (filp->f_mode & FMODE_READ) {
 526		mask = (nrbufs > 0) ? EPOLLIN | EPOLLRDNORM : 0;
 527		if (!pipe->writers && filp->f_version != pipe->w_counter)
 528			mask |= EPOLLHUP;
 529	}
 530
 531	if (filp->f_mode & FMODE_WRITE) {
 532		mask |= (nrbufs < pipe->buffers) ? EPOLLOUT | EPOLLWRNORM : 0;
 533		/*
 534		 * Most Unices do not set EPOLLERR for FIFOs but on Linux they
 535		 * behave exactly like pipes for poll().
 536		 */
 537		if (!pipe->readers)
 538			mask |= EPOLLERR;
 539	}
 540
 541	return mask;
 542}
 543
 544static void put_pipe_info(struct inode *inode, struct pipe_inode_info *pipe)
 545{
 546	int kill = 0;
 547
 548	spin_lock(&inode->i_lock);
 549	if (!--pipe->files) {
 550		inode->i_pipe = NULL;
 551		kill = 1;
 552	}
 553	spin_unlock(&inode->i_lock);
 554
 555	if (kill)
 556		free_pipe_info(pipe);
 557}
 558
 559static int
 560pipe_release(struct inode *inode, struct file *file)
 561{
 562	struct pipe_inode_info *pipe = file->private_data;
 563
 564	__pipe_lock(pipe);
 565	if (file->f_mode & FMODE_READ)
 566		pipe->readers--;
 567	if (file->f_mode & FMODE_WRITE)
 568		pipe->writers--;
 569
 570	if (pipe->readers || pipe->writers) {
 571		wake_up_interruptible_sync_poll(&pipe->wait, EPOLLIN | EPOLLOUT | EPOLLRDNORM | EPOLLWRNORM | EPOLLERR | EPOLLHUP);
 572		kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
 573		kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
 574	}
 575	__pipe_unlock(pipe);
 576
 577	put_pipe_info(inode, pipe);
 578	return 0;
 579}
 580
 581static int
 582pipe_fasync(int fd, struct file *filp, int on)
 583{
 584	struct pipe_inode_info *pipe = filp->private_data;
 585	int retval = 0;
 586
 587	__pipe_lock(pipe);
 588	if (filp->f_mode & FMODE_READ)
 589		retval = fasync_helper(fd, filp, on, &pipe->fasync_readers);
 590	if ((filp->f_mode & FMODE_WRITE) && retval >= 0) {
 591		retval = fasync_helper(fd, filp, on, &pipe->fasync_writers);
 592		if (retval < 0 && (filp->f_mode & FMODE_READ))
 593			/* this can happen only if on == T */
 594			fasync_helper(-1, filp, 0, &pipe->fasync_readers);
 595	}
 596	__pipe_unlock(pipe);
 597	return retval;
 598}
 599
 600static unsigned long account_pipe_buffers(struct user_struct *user,
 601                                 unsigned long old, unsigned long new)
 602{
 603	return atomic_long_add_return(new - old, &user->pipe_bufs);
 604}
 605
 606static bool too_many_pipe_buffers_soft(unsigned long user_bufs)
 607{
 608	unsigned long soft_limit = READ_ONCE(pipe_user_pages_soft);
 609
 610	return soft_limit && user_bufs > soft_limit;
 611}
 612
 613static bool too_many_pipe_buffers_hard(unsigned long user_bufs)
 614{
 615	unsigned long hard_limit = READ_ONCE(pipe_user_pages_hard);
 616
 617	return hard_limit && user_bufs > hard_limit;
 618}
 619
 620static bool is_unprivileged_user(void)
 621{
 622	return !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN);
 623}
 624
 625struct pipe_inode_info *alloc_pipe_info(void)
 626{
 627	struct pipe_inode_info *pipe;
 628	unsigned long pipe_bufs = PIPE_DEF_BUFFERS;
 629	struct user_struct *user = get_current_user();
 630	unsigned long user_bufs;
 631	unsigned int max_size = READ_ONCE(pipe_max_size);
 632
 633	pipe = kzalloc(sizeof(struct pipe_inode_info), GFP_KERNEL_ACCOUNT);
 634	if (pipe == NULL)
 635		goto out_free_uid;
 636
 637	if (pipe_bufs * PAGE_SIZE > max_size && !capable(CAP_SYS_RESOURCE))
 638		pipe_bufs = max_size >> PAGE_SHIFT;
 639
 640	user_bufs = account_pipe_buffers(user, 0, pipe_bufs);
 641
 642	if (too_many_pipe_buffers_soft(user_bufs) && is_unprivileged_user()) {
 643		user_bufs = account_pipe_buffers(user, pipe_bufs, 1);
 644		pipe_bufs = 1;
 645	}
 646
 647	if (too_many_pipe_buffers_hard(user_bufs) && is_unprivileged_user())
 648		goto out_revert_acct;
 649
 650	pipe->bufs = kcalloc(pipe_bufs, sizeof(struct pipe_buffer),
 651			     GFP_KERNEL_ACCOUNT);
 652
 653	if (pipe->bufs) {
 654		init_waitqueue_head(&pipe->wait);
 655		pipe->r_counter = pipe->w_counter = 1;
 656		pipe->buffers = pipe_bufs;
 657		pipe->user = user;
 658		mutex_init(&pipe->mutex);
 659		return pipe;
 
 
 
 
 660	}
 661
 662out_revert_acct:
 663	(void) account_pipe_buffers(user, pipe_bufs, 0);
 664	kfree(pipe);
 665out_free_uid:
 666	free_uid(user);
 667	return NULL;
 668}
 669
 670void free_pipe_info(struct pipe_inode_info *pipe)
 671{
 672	int i;
 673
 674	(void) account_pipe_buffers(pipe->user, pipe->buffers, 0);
 675	free_uid(pipe->user);
 676	for (i = 0; i < pipe->buffers; i++) {
 677		struct pipe_buffer *buf = pipe->bufs + i;
 678		if (buf->ops)
 679			pipe_buf_release(pipe, buf);
 680	}
 681	if (pipe->tmp_page)
 682		__free_page(pipe->tmp_page);
 683	kfree(pipe->bufs);
 684	kfree(pipe);
 685}
 686
 687static struct vfsmount *pipe_mnt __read_mostly;
 688
 689/*
 690 * pipefs_dname() is called from d_path().
 691 */
 692static char *pipefs_dname(struct dentry *dentry, char *buffer, int buflen)
 693{
 694	return dynamic_dname(dentry, buffer, buflen, "pipe:[%lu]",
 695				d_inode(dentry)->i_ino);
 696}
 697
 698static const struct dentry_operations pipefs_dentry_operations = {
 699	.d_dname	= pipefs_dname,
 700};
 701
 702static struct inode * get_pipe_inode(void)
 703{
 704	struct inode *inode = new_inode_pseudo(pipe_mnt->mnt_sb);
 705	struct pipe_inode_info *pipe;
 706
 707	if (!inode)
 708		goto fail_inode;
 709
 710	inode->i_ino = get_next_ino();
 711
 712	pipe = alloc_pipe_info();
 713	if (!pipe)
 714		goto fail_iput;
 715
 716	inode->i_pipe = pipe;
 717	pipe->files = 2;
 718	pipe->readers = pipe->writers = 1;
 719	inode->i_fop = &pipefifo_fops;
 720
 721	/*
 722	 * Mark the inode dirty from the very beginning,
 723	 * that way it will never be moved to the dirty
 724	 * list because "mark_inode_dirty()" will think
 725	 * that it already _is_ on the dirty list.
 726	 */
 727	inode->i_state = I_DIRTY;
 728	inode->i_mode = S_IFIFO | S_IRUSR | S_IWUSR;
 729	inode->i_uid = current_fsuid();
 730	inode->i_gid = current_fsgid();
 731	inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode);
 732
 733	return inode;
 734
 735fail_iput:
 736	iput(inode);
 737
 738fail_inode:
 739	return NULL;
 740}
 741
 742int create_pipe_files(struct file **res, int flags)
 743{
 744	int err;
 745	struct inode *inode = get_pipe_inode();
 746	struct file *f;
 747	struct path path;
 
 748
 749	if (!inode)
 750		return -ENFILE;
 751
 752	err = -ENOMEM;
 753	path.dentry = d_alloc_pseudo(pipe_mnt->mnt_sb, &empty_name);
 754	if (!path.dentry)
 755		goto err_inode;
 756	path.mnt = mntget(pipe_mnt);
 757
 758	d_instantiate(path.dentry, inode);
 759
 760	f = alloc_file(&path, FMODE_WRITE, &pipefifo_fops);
 761	if (IS_ERR(f)) {
 762		err = PTR_ERR(f);
 763		goto err_dentry;
 764	}
 765
 766	f->f_flags = O_WRONLY | (flags & (O_NONBLOCK | O_DIRECT));
 767	f->private_data = inode->i_pipe;
 768
 769	res[0] = alloc_file(&path, FMODE_READ, &pipefifo_fops);
 770	if (IS_ERR(res[0])) {
 771		err = PTR_ERR(res[0]);
 772		goto err_file;
 773	}
 774
 775	path_get(&path);
 776	res[0]->private_data = inode->i_pipe;
 777	res[0]->f_flags = O_RDONLY | (flags & O_NONBLOCK);
 778	res[1] = f;
 779	return 0;
 780
 781err_file:
 782	put_filp(f);
 783err_dentry:
 784	free_pipe_info(inode->i_pipe);
 785	path_put(&path);
 786	return err;
 787
 788err_inode:
 789	free_pipe_info(inode->i_pipe);
 790	iput(inode);
 791	return err;
 792}
 793
 794static int __do_pipe_flags(int *fd, struct file **files, int flags)
 795{
 796	int error;
 797	int fdw, fdr;
 798
 799	if (flags & ~(O_CLOEXEC | O_NONBLOCK | O_DIRECT))
 800		return -EINVAL;
 801
 802	error = create_pipe_files(files, flags);
 803	if (error)
 804		return error;
 805
 806	error = get_unused_fd_flags(flags);
 807	if (error < 0)
 808		goto err_read_pipe;
 809	fdr = error;
 810
 811	error = get_unused_fd_flags(flags);
 812	if (error < 0)
 813		goto err_fdr;
 814	fdw = error;
 815
 816	audit_fd_pair(fdr, fdw);
 817	fd[0] = fdr;
 818	fd[1] = fdw;
 819	return 0;
 820
 821 err_fdr:
 822	put_unused_fd(fdr);
 823 err_read_pipe:
 824	fput(files[0]);
 825	fput(files[1]);
 826	return error;
 827}
 828
 829int do_pipe_flags(int *fd, int flags)
 830{
 831	struct file *files[2];
 832	int error = __do_pipe_flags(fd, files, flags);
 833	if (!error) {
 834		fd_install(fd[0], files[0]);
 835		fd_install(fd[1], files[1]);
 836	}
 837	return error;
 838}
 839
 840/*
 841 * sys_pipe() is the normal C calling standard for creating
 842 * a pipe. It's not the way Unix traditionally does this, though.
 843 */
 844static int do_pipe2(int __user *fildes, int flags)
 845{
 846	struct file *files[2];
 847	int fd[2];
 848	int error;
 849
 850	error = __do_pipe_flags(fd, files, flags);
 851	if (!error) {
 852		if (unlikely(copy_to_user(fildes, fd, sizeof(fd)))) {
 853			fput(files[0]);
 854			fput(files[1]);
 855			put_unused_fd(fd[0]);
 856			put_unused_fd(fd[1]);
 857			error = -EFAULT;
 858		} else {
 859			fd_install(fd[0], files[0]);
 860			fd_install(fd[1], files[1]);
 861		}
 862	}
 863	return error;
 864}
 865
 866SYSCALL_DEFINE2(pipe2, int __user *, fildes, int, flags)
 867{
 868	return do_pipe2(fildes, flags);
 869}
 870
 871SYSCALL_DEFINE1(pipe, int __user *, fildes)
 872{
 873	return do_pipe2(fildes, 0);
 874}
 875
 876static int wait_for_partner(struct pipe_inode_info *pipe, unsigned int *cnt)
 877{
 878	int cur = *cnt;	
 879
 880	while (cur == *cnt) {
 881		pipe_wait(pipe);
 882		if (signal_pending(current))
 883			break;
 884	}
 885	return cur == *cnt ? -ERESTARTSYS : 0;
 886}
 887
 888static void wake_up_partner(struct pipe_inode_info *pipe)
 889{
 890	wake_up_interruptible(&pipe->wait);
 891}
 892
 893static int fifo_open(struct inode *inode, struct file *filp)
 894{
 895	struct pipe_inode_info *pipe;
 896	bool is_pipe = inode->i_sb->s_magic == PIPEFS_MAGIC;
 897	int ret;
 898
 899	filp->f_version = 0;
 900
 901	spin_lock(&inode->i_lock);
 902	if (inode->i_pipe) {
 903		pipe = inode->i_pipe;
 904		pipe->files++;
 905		spin_unlock(&inode->i_lock);
 906	} else {
 907		spin_unlock(&inode->i_lock);
 908		pipe = alloc_pipe_info();
 909		if (!pipe)
 910			return -ENOMEM;
 911		pipe->files = 1;
 912		spin_lock(&inode->i_lock);
 913		if (unlikely(inode->i_pipe)) {
 914			inode->i_pipe->files++;
 915			spin_unlock(&inode->i_lock);
 916			free_pipe_info(pipe);
 917			pipe = inode->i_pipe;
 918		} else {
 919			inode->i_pipe = pipe;
 920			spin_unlock(&inode->i_lock);
 921		}
 922	}
 923	filp->private_data = pipe;
 924	/* OK, we have a pipe and it's pinned down */
 925
 926	__pipe_lock(pipe);
 927
 928	/* We can only do regular read/write on fifos */
 929	filp->f_mode &= (FMODE_READ | FMODE_WRITE);
 930
 931	switch (filp->f_mode) {
 932	case FMODE_READ:
 933	/*
 934	 *  O_RDONLY
 935	 *  POSIX.1 says that O_NONBLOCK means return with the FIFO
 936	 *  opened, even when there is no process writing the FIFO.
 937	 */
 938		pipe->r_counter++;
 939		if (pipe->readers++ == 0)
 940			wake_up_partner(pipe);
 941
 942		if (!is_pipe && !pipe->writers) {
 943			if ((filp->f_flags & O_NONBLOCK)) {
 944				/* suppress EPOLLHUP until we have
 945				 * seen a writer */
 946				filp->f_version = pipe->w_counter;
 947			} else {
 948				if (wait_for_partner(pipe, &pipe->w_counter))
 949					goto err_rd;
 950			}
 951		}
 952		break;
 953	
 954	case FMODE_WRITE:
 955	/*
 956	 *  O_WRONLY
 957	 *  POSIX.1 says that O_NONBLOCK means return -1 with
 958	 *  errno=ENXIO when there is no process reading the FIFO.
 959	 */
 960		ret = -ENXIO;
 961		if (!is_pipe && (filp->f_flags & O_NONBLOCK) && !pipe->readers)
 962			goto err;
 963
 964		pipe->w_counter++;
 965		if (!pipe->writers++)
 966			wake_up_partner(pipe);
 967
 968		if (!is_pipe && !pipe->readers) {
 969			if (wait_for_partner(pipe, &pipe->r_counter))
 970				goto err_wr;
 971		}
 972		break;
 973	
 974	case FMODE_READ | FMODE_WRITE:
 975	/*
 976	 *  O_RDWR
 977	 *  POSIX.1 leaves this case "undefined" when O_NONBLOCK is set.
 978	 *  This implementation will NEVER block on a O_RDWR open, since
 979	 *  the process can at least talk to itself.
 980	 */
 981
 982		pipe->readers++;
 983		pipe->writers++;
 984		pipe->r_counter++;
 985		pipe->w_counter++;
 986		if (pipe->readers == 1 || pipe->writers == 1)
 987			wake_up_partner(pipe);
 988		break;
 989
 990	default:
 991		ret = -EINVAL;
 992		goto err;
 993	}
 994
 995	/* Ok! */
 996	__pipe_unlock(pipe);
 997	return 0;
 998
 999err_rd:
1000	if (!--pipe->readers)
1001		wake_up_interruptible(&pipe->wait);
1002	ret = -ERESTARTSYS;
1003	goto err;
1004
1005err_wr:
1006	if (!--pipe->writers)
1007		wake_up_interruptible(&pipe->wait);
1008	ret = -ERESTARTSYS;
1009	goto err;
1010
1011err:
1012	__pipe_unlock(pipe);
1013
1014	put_pipe_info(inode, pipe);
1015	return ret;
1016}
1017
1018const struct file_operations pipefifo_fops = {
1019	.open		= fifo_open,
1020	.llseek		= no_llseek,
1021	.read_iter	= pipe_read,
1022	.write_iter	= pipe_write,
1023	.poll		= pipe_poll,
1024	.unlocked_ioctl	= pipe_ioctl,
1025	.release	= pipe_release,
1026	.fasync		= pipe_fasync,
1027};
1028
1029/*
1030 * Currently we rely on the pipe array holding a power-of-2 number
1031 * of pages. Returns 0 on error.
1032 */
1033unsigned int round_pipe_size(unsigned long size)
1034{
1035	if (size > (1U << 31))
1036		return 0;
1037
1038	/* Minimum pipe size, as required by POSIX */
1039	if (size < PAGE_SIZE)
1040		return PAGE_SIZE;
1041
1042	return roundup_pow_of_two(size);
1043}
1044
1045/*
1046 * Allocate a new array of pipe buffers and copy the info over. Returns the
1047 * pipe size if successful, or return -ERROR on error.
1048 */
1049static long pipe_set_size(struct pipe_inode_info *pipe, unsigned long arg)
1050{
1051	struct pipe_buffer *bufs;
1052	unsigned int size, nr_pages;
1053	unsigned long user_bufs;
1054	long ret = 0;
1055
1056	size = round_pipe_size(arg);
1057	nr_pages = size >> PAGE_SHIFT;
1058
1059	if (!nr_pages)
1060		return -EINVAL;
1061
1062	/*
1063	 * If trying to increase the pipe capacity, check that an
1064	 * unprivileged user is not trying to exceed various limits
1065	 * (soft limit check here, hard limit check just below).
1066	 * Decreasing the pipe capacity is always permitted, even
1067	 * if the user is currently over a limit.
1068	 */
1069	if (nr_pages > pipe->buffers &&
1070			size > pipe_max_size && !capable(CAP_SYS_RESOURCE))
1071		return -EPERM;
1072
1073	user_bufs = account_pipe_buffers(pipe->user, pipe->buffers, nr_pages);
1074
1075	if (nr_pages > pipe->buffers &&
1076			(too_many_pipe_buffers_hard(user_bufs) ||
1077			 too_many_pipe_buffers_soft(user_bufs)) &&
1078			is_unprivileged_user()) {
1079		ret = -EPERM;
1080		goto out_revert_acct;
1081	}
1082
1083	/*
1084	 * We can shrink the pipe, if arg >= pipe->nrbufs. Since we don't
1085	 * expect a lot of shrink+grow operations, just free and allocate
1086	 * again like we would do for growing. If the pipe currently
1087	 * contains more buffers than arg, then return busy.
1088	 */
1089	if (nr_pages < pipe->nrbufs) {
1090		ret = -EBUSY;
1091		goto out_revert_acct;
1092	}
1093
1094	bufs = kcalloc(nr_pages, sizeof(*bufs),
1095		       GFP_KERNEL_ACCOUNT | __GFP_NOWARN);
1096	if (unlikely(!bufs)) {
1097		ret = -ENOMEM;
1098		goto out_revert_acct;
1099	}
1100
1101	/*
1102	 * The pipe array wraps around, so just start the new one at zero
1103	 * and adjust the indexes.
1104	 */
1105	if (pipe->nrbufs) {
1106		unsigned int tail;
1107		unsigned int head;
1108
1109		tail = pipe->curbuf + pipe->nrbufs;
1110		if (tail < pipe->buffers)
1111			tail = 0;
1112		else
1113			tail &= (pipe->buffers - 1);
1114
1115		head = pipe->nrbufs - tail;
1116		if (head)
1117			memcpy(bufs, pipe->bufs + pipe->curbuf, head * sizeof(struct pipe_buffer));
1118		if (tail)
1119			memcpy(bufs + head, pipe->bufs, tail * sizeof(struct pipe_buffer));
1120	}
1121
 
1122	pipe->curbuf = 0;
1123	kfree(pipe->bufs);
1124	pipe->bufs = bufs;
1125	pipe->buffers = nr_pages;
1126	return nr_pages * PAGE_SIZE;
 
1127
1128out_revert_acct:
1129	(void) account_pipe_buffers(pipe->user, nr_pages, pipe->buffers);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1130	return ret;
1131}
1132
1133/*
1134 * After the inode slimming patch, i_pipe/i_bdev/i_cdev share the same
1135 * location, so checking ->i_pipe is not enough to verify that this is a
1136 * pipe.
1137 */
1138struct pipe_inode_info *get_pipe_info(struct file *file)
1139{
1140	return file->f_op == &pipefifo_fops ? file->private_data : NULL;
1141}
1142
1143long pipe_fcntl(struct file *file, unsigned int cmd, unsigned long arg)
1144{
1145	struct pipe_inode_info *pipe;
1146	long ret;
1147
1148	pipe = get_pipe_info(file);
1149	if (!pipe)
1150		return -EBADF;
1151
1152	__pipe_lock(pipe);
1153
1154	switch (cmd) {
1155	case F_SETPIPE_SZ:
1156		ret = pipe_set_size(pipe, arg);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1157		break;
 
1158	case F_GETPIPE_SZ:
1159		ret = pipe->buffers * PAGE_SIZE;
1160		break;
1161	default:
1162		ret = -EINVAL;
1163		break;
1164	}
1165
 
1166	__pipe_unlock(pipe);
1167	return ret;
1168}
1169
1170static const struct super_operations pipefs_ops = {
1171	.destroy_inode = free_inode_nonrcu,
1172	.statfs = simple_statfs,
1173};
1174
1175/*
1176 * pipefs should _never_ be mounted by userland - too much of security hassle,
1177 * no real gain from having the whole whorehouse mounted. So we don't need
1178 * any operations on the root directory. However, we need a non-trivial
1179 * d_name - pipe: will go nicely and kill the special-casing in procfs.
1180 */
1181static struct dentry *pipefs_mount(struct file_system_type *fs_type,
1182			 int flags, const char *dev_name, void *data)
1183{
1184	return mount_pseudo(fs_type, "pipe:", &pipefs_ops,
1185			&pipefs_dentry_operations, PIPEFS_MAGIC);
1186}
1187
1188static struct file_system_type pipe_fs_type = {
1189	.name		= "pipefs",
1190	.mount		= pipefs_mount,
1191	.kill_sb	= kill_anon_super,
1192};
1193
1194static int __init init_pipe_fs(void)
1195{
1196	int err = register_filesystem(&pipe_fs_type);
1197
1198	if (!err) {
1199		pipe_mnt = kern_mount(&pipe_fs_type);
1200		if (IS_ERR(pipe_mnt)) {
1201			err = PTR_ERR(pipe_mnt);
1202			unregister_filesystem(&pipe_fs_type);
1203		}
1204	}
1205	return err;
1206}
1207
1208fs_initcall(init_pipe_fs);
v4.6
 
   1/*
   2 *  linux/fs/pipe.c
   3 *
   4 *  Copyright (C) 1991, 1992, 1999  Linus Torvalds
   5 */
   6
   7#include <linux/mm.h>
   8#include <linux/file.h>
   9#include <linux/poll.h>
  10#include <linux/slab.h>
  11#include <linux/module.h>
  12#include <linux/init.h>
  13#include <linux/fs.h>
  14#include <linux/log2.h>
  15#include <linux/mount.h>
  16#include <linux/magic.h>
  17#include <linux/pipe_fs_i.h>
  18#include <linux/uio.h>
  19#include <linux/highmem.h>
  20#include <linux/pagemap.h>
  21#include <linux/audit.h>
  22#include <linux/syscalls.h>
  23#include <linux/fcntl.h>
 
  24
  25#include <asm/uaccess.h>
  26#include <asm/ioctls.h>
  27
  28#include "internal.h"
  29
  30/*
  31 * The max size that a non-root user is allowed to grow the pipe. Can
  32 * be set by root in /proc/sys/fs/pipe-max-size
  33 */
  34unsigned int pipe_max_size = 1048576;
  35
  36/*
  37 * Minimum pipe size, as required by POSIX
  38 */
  39unsigned int pipe_min_size = PAGE_SIZE;
  40
  41/* Maximum allocatable pages per user. Hard limit is unset by default, soft
  42 * matches default values.
  43 */
  44unsigned long pipe_user_pages_hard;
  45unsigned long pipe_user_pages_soft = PIPE_DEF_BUFFERS * INR_OPEN_CUR;
  46
  47/*
  48 * We use a start+len construction, which provides full use of the 
  49 * allocated memory.
  50 * -- Florian Coosmann (FGC)
  51 * 
  52 * Reads with count = 0 should always return 0.
  53 * -- Julian Bradfield 1999-06-07.
  54 *
  55 * FIFOs and Pipes now generate SIGIO for both readers and writers.
  56 * -- Jeremy Elson <jelson@circlemud.org> 2001-08-16
  57 *
  58 * pipe_read & write cleanup
  59 * -- Manfred Spraul <manfred@colorfullife.com> 2002-05-09
  60 */
  61
  62static void pipe_lock_nested(struct pipe_inode_info *pipe, int subclass)
  63{
  64	if (pipe->files)
  65		mutex_lock_nested(&pipe->mutex, subclass);
  66}
  67
  68void pipe_lock(struct pipe_inode_info *pipe)
  69{
  70	/*
  71	 * pipe_lock() nests non-pipe inode locks (for writing to a file)
  72	 */
  73	pipe_lock_nested(pipe, I_MUTEX_PARENT);
  74}
  75EXPORT_SYMBOL(pipe_lock);
  76
  77void pipe_unlock(struct pipe_inode_info *pipe)
  78{
  79	if (pipe->files)
  80		mutex_unlock(&pipe->mutex);
  81}
  82EXPORT_SYMBOL(pipe_unlock);
  83
  84static inline void __pipe_lock(struct pipe_inode_info *pipe)
  85{
  86	mutex_lock_nested(&pipe->mutex, I_MUTEX_PARENT);
  87}
  88
  89static inline void __pipe_unlock(struct pipe_inode_info *pipe)
  90{
  91	mutex_unlock(&pipe->mutex);
  92}
  93
  94void pipe_double_lock(struct pipe_inode_info *pipe1,
  95		      struct pipe_inode_info *pipe2)
  96{
  97	BUG_ON(pipe1 == pipe2);
  98
  99	if (pipe1 < pipe2) {
 100		pipe_lock_nested(pipe1, I_MUTEX_PARENT);
 101		pipe_lock_nested(pipe2, I_MUTEX_CHILD);
 102	} else {
 103		pipe_lock_nested(pipe2, I_MUTEX_PARENT);
 104		pipe_lock_nested(pipe1, I_MUTEX_CHILD);
 105	}
 106}
 107
 108/* Drop the inode semaphore and wait for a pipe event, atomically */
 109void pipe_wait(struct pipe_inode_info *pipe)
 110{
 111	DEFINE_WAIT(wait);
 112
 113	/*
 114	 * Pipes are system-local resources, so sleeping on them
 115	 * is considered a noninteractive wait:
 116	 */
 117	prepare_to_wait(&pipe->wait, &wait, TASK_INTERRUPTIBLE);
 118	pipe_unlock(pipe);
 119	schedule();
 120	finish_wait(&pipe->wait, &wait);
 121	pipe_lock(pipe);
 122}
 123
 124static void anon_pipe_buf_release(struct pipe_inode_info *pipe,
 125				  struct pipe_buffer *buf)
 126{
 127	struct page *page = buf->page;
 128
 129	/*
 130	 * If nobody else uses this page, and we don't already have a
 131	 * temporary page, let's keep track of it as a one-deep
 132	 * allocation cache. (Otherwise just release our reference to it)
 133	 */
 134	if (page_count(page) == 1 && !pipe->tmp_page)
 135		pipe->tmp_page = page;
 136	else
 137		put_page(page);
 138}
 139
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 140/**
 141 * generic_pipe_buf_steal - attempt to take ownership of a &pipe_buffer
 142 * @pipe:	the pipe that the buffer belongs to
 143 * @buf:	the buffer to attempt to steal
 144 *
 145 * Description:
 146 *	This function attempts to steal the &struct page attached to
 147 *	@buf. If successful, this function returns 0 and returns with
 148 *	the page locked. The caller may then reuse the page for whatever
 149 *	he wishes; the typical use is insertion into a different file
 150 *	page cache.
 151 */
 152int generic_pipe_buf_steal(struct pipe_inode_info *pipe,
 153			   struct pipe_buffer *buf)
 154{
 155	struct page *page = buf->page;
 156
 157	/*
 158	 * A reference of one is golden, that means that the owner of this
 159	 * page is the only one holding a reference to it. lock the page
 160	 * and return OK.
 161	 */
 162	if (page_count(page) == 1) {
 163		lock_page(page);
 164		return 0;
 165	}
 166
 167	return 1;
 168}
 169EXPORT_SYMBOL(generic_pipe_buf_steal);
 170
 171/**
 172 * generic_pipe_buf_get - get a reference to a &struct pipe_buffer
 173 * @pipe:	the pipe that the buffer belongs to
 174 * @buf:	the buffer to get a reference to
 175 *
 176 * Description:
 177 *	This function grabs an extra reference to @buf. It's used in
 178 *	in the tee() system call, when we duplicate the buffers in one
 179 *	pipe into another.
 180 */
 181void generic_pipe_buf_get(struct pipe_inode_info *pipe, struct pipe_buffer *buf)
 182{
 183	get_page(buf->page);
 184}
 185EXPORT_SYMBOL(generic_pipe_buf_get);
 186
 187/**
 188 * generic_pipe_buf_confirm - verify contents of the pipe buffer
 189 * @info:	the pipe that the buffer belongs to
 190 * @buf:	the buffer to confirm
 191 *
 192 * Description:
 193 *	This function does nothing, because the generic pipe code uses
 194 *	pages that are always good when inserted into the pipe.
 195 */
 196int generic_pipe_buf_confirm(struct pipe_inode_info *info,
 197			     struct pipe_buffer *buf)
 198{
 199	return 0;
 200}
 201EXPORT_SYMBOL(generic_pipe_buf_confirm);
 202
 203/**
 204 * generic_pipe_buf_release - put a reference to a &struct pipe_buffer
 205 * @pipe:	the pipe that the buffer belongs to
 206 * @buf:	the buffer to put a reference to
 207 *
 208 * Description:
 209 *	This function releases a reference to @buf.
 210 */
 211void generic_pipe_buf_release(struct pipe_inode_info *pipe,
 212			      struct pipe_buffer *buf)
 213{
 214	put_page(buf->page);
 215}
 216EXPORT_SYMBOL(generic_pipe_buf_release);
 217
 218static const struct pipe_buf_operations anon_pipe_buf_ops = {
 219	.can_merge = 1,
 220	.confirm = generic_pipe_buf_confirm,
 221	.release = anon_pipe_buf_release,
 222	.steal = generic_pipe_buf_steal,
 223	.get = generic_pipe_buf_get,
 224};
 225
 226static const struct pipe_buf_operations packet_pipe_buf_ops = {
 227	.can_merge = 0,
 228	.confirm = generic_pipe_buf_confirm,
 229	.release = anon_pipe_buf_release,
 230	.steal = generic_pipe_buf_steal,
 231	.get = generic_pipe_buf_get,
 232};
 233
 234static ssize_t
 235pipe_read(struct kiocb *iocb, struct iov_iter *to)
 236{
 237	size_t total_len = iov_iter_count(to);
 238	struct file *filp = iocb->ki_filp;
 239	struct pipe_inode_info *pipe = filp->private_data;
 240	int do_wakeup;
 241	ssize_t ret;
 242
 243	/* Null read succeeds. */
 244	if (unlikely(total_len == 0))
 245		return 0;
 246
 247	do_wakeup = 0;
 248	ret = 0;
 249	__pipe_lock(pipe);
 250	for (;;) {
 251		int bufs = pipe->nrbufs;
 252		if (bufs) {
 253			int curbuf = pipe->curbuf;
 254			struct pipe_buffer *buf = pipe->bufs + curbuf;
 255			const struct pipe_buf_operations *ops = buf->ops;
 256			size_t chars = buf->len;
 257			size_t written;
 258			int error;
 259
 260			if (chars > total_len)
 261				chars = total_len;
 262
 263			error = ops->confirm(pipe, buf);
 264			if (error) {
 265				if (!ret)
 266					ret = error;
 267				break;
 268			}
 269
 270			written = copy_page_to_iter(buf->page, buf->offset, chars, to);
 271			if (unlikely(written < chars)) {
 272				if (!ret)
 273					ret = -EFAULT;
 274				break;
 275			}
 276			ret += chars;
 277			buf->offset += chars;
 278			buf->len -= chars;
 279
 280			/* Was it a packet buffer? Clean up and exit */
 281			if (buf->flags & PIPE_BUF_FLAG_PACKET) {
 282				total_len = chars;
 283				buf->len = 0;
 284			}
 285
 286			if (!buf->len) {
 287				buf->ops = NULL;
 288				ops->release(pipe, buf);
 289				curbuf = (curbuf + 1) & (pipe->buffers - 1);
 290				pipe->curbuf = curbuf;
 291				pipe->nrbufs = --bufs;
 292				do_wakeup = 1;
 293			}
 294			total_len -= chars;
 295			if (!total_len)
 296				break;	/* common path: read succeeded */
 297		}
 298		if (bufs)	/* More to do? */
 299			continue;
 300		if (!pipe->writers)
 301			break;
 302		if (!pipe->waiting_writers) {
 303			/* syscall merging: Usually we must not sleep
 304			 * if O_NONBLOCK is set, or if we got some data.
 305			 * But if a writer sleeps in kernel space, then
 306			 * we can wait for that data without violating POSIX.
 307			 */
 308			if (ret)
 309				break;
 310			if (filp->f_flags & O_NONBLOCK) {
 311				ret = -EAGAIN;
 312				break;
 313			}
 314		}
 315		if (signal_pending(current)) {
 316			if (!ret)
 317				ret = -ERESTARTSYS;
 318			break;
 319		}
 320		if (do_wakeup) {
 321			wake_up_interruptible_sync_poll(&pipe->wait, POLLOUT | POLLWRNORM);
 322 			kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
 323		}
 324		pipe_wait(pipe);
 325	}
 326	__pipe_unlock(pipe);
 327
 328	/* Signal writers asynchronously that there is more room. */
 329	if (do_wakeup) {
 330		wake_up_interruptible_sync_poll(&pipe->wait, POLLOUT | POLLWRNORM);
 331		kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
 332	}
 333	if (ret > 0)
 334		file_accessed(filp);
 335	return ret;
 336}
 337
 338static inline int is_packetized(struct file *file)
 339{
 340	return (file->f_flags & O_DIRECT) != 0;
 341}
 342
 343static ssize_t
 344pipe_write(struct kiocb *iocb, struct iov_iter *from)
 345{
 346	struct file *filp = iocb->ki_filp;
 347	struct pipe_inode_info *pipe = filp->private_data;
 348	ssize_t ret = 0;
 349	int do_wakeup = 0;
 350	size_t total_len = iov_iter_count(from);
 351	ssize_t chars;
 352
 353	/* Null write succeeds. */
 354	if (unlikely(total_len == 0))
 355		return 0;
 356
 357	__pipe_lock(pipe);
 358
 359	if (!pipe->readers) {
 360		send_sig(SIGPIPE, current, 0);
 361		ret = -EPIPE;
 362		goto out;
 363	}
 364
 365	/* We try to merge small writes */
 366	chars = total_len & (PAGE_SIZE-1); /* size of the last buffer */
 367	if (pipe->nrbufs && chars != 0) {
 368		int lastbuf = (pipe->curbuf + pipe->nrbufs - 1) &
 369							(pipe->buffers - 1);
 370		struct pipe_buffer *buf = pipe->bufs + lastbuf;
 371		const struct pipe_buf_operations *ops = buf->ops;
 372		int offset = buf->offset + buf->len;
 373
 374		if (ops->can_merge && offset + chars <= PAGE_SIZE) {
 375			ret = ops->confirm(pipe, buf);
 376			if (ret)
 377				goto out;
 378
 379			ret = copy_page_from_iter(buf->page, offset, chars, from);
 380			if (unlikely(ret < chars)) {
 381				ret = -EFAULT;
 382				goto out;
 383			}
 384			do_wakeup = 1;
 385			buf->len += ret;
 386			if (!iov_iter_count(from))
 387				goto out;
 388		}
 389	}
 390
 391	for (;;) {
 392		int bufs;
 393
 394		if (!pipe->readers) {
 395			send_sig(SIGPIPE, current, 0);
 396			if (!ret)
 397				ret = -EPIPE;
 398			break;
 399		}
 400		bufs = pipe->nrbufs;
 401		if (bufs < pipe->buffers) {
 402			int newbuf = (pipe->curbuf + bufs) & (pipe->buffers-1);
 403			struct pipe_buffer *buf = pipe->bufs + newbuf;
 404			struct page *page = pipe->tmp_page;
 405			int copied;
 406
 407			if (!page) {
 408				page = alloc_page(GFP_HIGHUSER);
 409				if (unlikely(!page)) {
 410					ret = ret ? : -ENOMEM;
 411					break;
 412				}
 413				pipe->tmp_page = page;
 414			}
 415			/* Always wake up, even if the copy fails. Otherwise
 416			 * we lock up (O_NONBLOCK-)readers that sleep due to
 417			 * syscall merging.
 418			 * FIXME! Is this really true?
 419			 */
 420			do_wakeup = 1;
 421			copied = copy_page_from_iter(page, 0, PAGE_SIZE, from);
 422			if (unlikely(copied < PAGE_SIZE && iov_iter_count(from))) {
 423				if (!ret)
 424					ret = -EFAULT;
 425				break;
 426			}
 427			ret += copied;
 428
 429			/* Insert it into the buffer array */
 430			buf->page = page;
 431			buf->ops = &anon_pipe_buf_ops;
 432			buf->offset = 0;
 433			buf->len = copied;
 434			buf->flags = 0;
 435			if (is_packetized(filp)) {
 436				buf->ops = &packet_pipe_buf_ops;
 437				buf->flags = PIPE_BUF_FLAG_PACKET;
 438			}
 439			pipe->nrbufs = ++bufs;
 440			pipe->tmp_page = NULL;
 441
 442			if (!iov_iter_count(from))
 443				break;
 444		}
 445		if (bufs < pipe->buffers)
 446			continue;
 447		if (filp->f_flags & O_NONBLOCK) {
 448			if (!ret)
 449				ret = -EAGAIN;
 450			break;
 451		}
 452		if (signal_pending(current)) {
 453			if (!ret)
 454				ret = -ERESTARTSYS;
 455			break;
 456		}
 457		if (do_wakeup) {
 458			wake_up_interruptible_sync_poll(&pipe->wait, POLLIN | POLLRDNORM);
 459			kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
 460			do_wakeup = 0;
 461		}
 462		pipe->waiting_writers++;
 463		pipe_wait(pipe);
 464		pipe->waiting_writers--;
 465	}
 466out:
 467	__pipe_unlock(pipe);
 468	if (do_wakeup) {
 469		wake_up_interruptible_sync_poll(&pipe->wait, POLLIN | POLLRDNORM);
 470		kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
 471	}
 472	if (ret > 0 && sb_start_write_trylock(file_inode(filp)->i_sb)) {
 473		int err = file_update_time(filp);
 474		if (err)
 475			ret = err;
 476		sb_end_write(file_inode(filp)->i_sb);
 477	}
 478	return ret;
 479}
 480
 481static long pipe_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
 482{
 483	struct pipe_inode_info *pipe = filp->private_data;
 484	int count, buf, nrbufs;
 485
 486	switch (cmd) {
 487		case FIONREAD:
 488			__pipe_lock(pipe);
 489			count = 0;
 490			buf = pipe->curbuf;
 491			nrbufs = pipe->nrbufs;
 492			while (--nrbufs >= 0) {
 493				count += pipe->bufs[buf].len;
 494				buf = (buf+1) & (pipe->buffers - 1);
 495			}
 496			__pipe_unlock(pipe);
 497
 498			return put_user(count, (int __user *)arg);
 499		default:
 500			return -ENOIOCTLCMD;
 501	}
 502}
 503
 504/* No kernel lock held - fine */
 505static unsigned int
 506pipe_poll(struct file *filp, poll_table *wait)
 507{
 508	unsigned int mask;
 509	struct pipe_inode_info *pipe = filp->private_data;
 510	int nrbufs;
 511
 512	poll_wait(filp, &pipe->wait, wait);
 513
 514	/* Reading only -- no need for acquiring the semaphore.  */
 515	nrbufs = pipe->nrbufs;
 516	mask = 0;
 517	if (filp->f_mode & FMODE_READ) {
 518		mask = (nrbufs > 0) ? POLLIN | POLLRDNORM : 0;
 519		if (!pipe->writers && filp->f_version != pipe->w_counter)
 520			mask |= POLLHUP;
 521	}
 522
 523	if (filp->f_mode & FMODE_WRITE) {
 524		mask |= (nrbufs < pipe->buffers) ? POLLOUT | POLLWRNORM : 0;
 525		/*
 526		 * Most Unices do not set POLLERR for FIFOs but on Linux they
 527		 * behave exactly like pipes for poll().
 528		 */
 529		if (!pipe->readers)
 530			mask |= POLLERR;
 531	}
 532
 533	return mask;
 534}
 535
 536static void put_pipe_info(struct inode *inode, struct pipe_inode_info *pipe)
 537{
 538	int kill = 0;
 539
 540	spin_lock(&inode->i_lock);
 541	if (!--pipe->files) {
 542		inode->i_pipe = NULL;
 543		kill = 1;
 544	}
 545	spin_unlock(&inode->i_lock);
 546
 547	if (kill)
 548		free_pipe_info(pipe);
 549}
 550
 551static int
 552pipe_release(struct inode *inode, struct file *file)
 553{
 554	struct pipe_inode_info *pipe = file->private_data;
 555
 556	__pipe_lock(pipe);
 557	if (file->f_mode & FMODE_READ)
 558		pipe->readers--;
 559	if (file->f_mode & FMODE_WRITE)
 560		pipe->writers--;
 561
 562	if (pipe->readers || pipe->writers) {
 563		wake_up_interruptible_sync_poll(&pipe->wait, POLLIN | POLLOUT | POLLRDNORM | POLLWRNORM | POLLERR | POLLHUP);
 564		kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
 565		kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
 566	}
 567	__pipe_unlock(pipe);
 568
 569	put_pipe_info(inode, pipe);
 570	return 0;
 571}
 572
 573static int
 574pipe_fasync(int fd, struct file *filp, int on)
 575{
 576	struct pipe_inode_info *pipe = filp->private_data;
 577	int retval = 0;
 578
 579	__pipe_lock(pipe);
 580	if (filp->f_mode & FMODE_READ)
 581		retval = fasync_helper(fd, filp, on, &pipe->fasync_readers);
 582	if ((filp->f_mode & FMODE_WRITE) && retval >= 0) {
 583		retval = fasync_helper(fd, filp, on, &pipe->fasync_writers);
 584		if (retval < 0 && (filp->f_mode & FMODE_READ))
 585			/* this can happen only if on == T */
 586			fasync_helper(-1, filp, 0, &pipe->fasync_readers);
 587	}
 588	__pipe_unlock(pipe);
 589	return retval;
 590}
 591
 592static void account_pipe_buffers(struct pipe_inode_info *pipe,
 593                                 unsigned long old, unsigned long new)
 594{
 595	atomic_long_add(new - old, &pipe->user->pipe_bufs);
 596}
 597
 598static bool too_many_pipe_buffers_soft(struct user_struct *user)
 599{
 600	return pipe_user_pages_soft &&
 601	       atomic_long_read(&user->pipe_bufs) >= pipe_user_pages_soft;
 
 602}
 603
 604static bool too_many_pipe_buffers_hard(struct user_struct *user)
 605{
 606	return pipe_user_pages_hard &&
 607	       atomic_long_read(&user->pipe_bufs) >= pipe_user_pages_hard;
 
 
 
 
 
 
 608}
 609
 610struct pipe_inode_info *alloc_pipe_info(void)
 611{
 612	struct pipe_inode_info *pipe;
 
 
 
 
 
 
 
 
 613
 614	pipe = kzalloc(sizeof(struct pipe_inode_info), GFP_KERNEL);
 615	if (pipe) {
 616		unsigned long pipe_bufs = PIPE_DEF_BUFFERS;
 617		struct user_struct *user = get_current_user();
 618
 619		if (!too_many_pipe_buffers_hard(user)) {
 620			if (too_many_pipe_buffers_soft(user))
 621				pipe_bufs = 1;
 622			pipe->bufs = kzalloc(sizeof(struct pipe_buffer) * pipe_bufs, GFP_KERNEL);
 623		}
 
 
 
 
 
 624
 625		if (pipe->bufs) {
 626			init_waitqueue_head(&pipe->wait);
 627			pipe->r_counter = pipe->w_counter = 1;
 628			pipe->buffers = pipe_bufs;
 629			pipe->user = user;
 630			account_pipe_buffers(pipe, 0, pipe_bufs);
 631			mutex_init(&pipe->mutex);
 632			return pipe;
 633		}
 634		free_uid(user);
 635		kfree(pipe);
 636	}
 637
 
 
 
 
 
 638	return NULL;
 639}
 640
 641void free_pipe_info(struct pipe_inode_info *pipe)
 642{
 643	int i;
 644
 645	account_pipe_buffers(pipe, pipe->buffers, 0);
 646	free_uid(pipe->user);
 647	for (i = 0; i < pipe->buffers; i++) {
 648		struct pipe_buffer *buf = pipe->bufs + i;
 649		if (buf->ops)
 650			buf->ops->release(pipe, buf);
 651	}
 652	if (pipe->tmp_page)
 653		__free_page(pipe->tmp_page);
 654	kfree(pipe->bufs);
 655	kfree(pipe);
 656}
 657
 658static struct vfsmount *pipe_mnt __read_mostly;
 659
 660/*
 661 * pipefs_dname() is called from d_path().
 662 */
 663static char *pipefs_dname(struct dentry *dentry, char *buffer, int buflen)
 664{
 665	return dynamic_dname(dentry, buffer, buflen, "pipe:[%lu]",
 666				d_inode(dentry)->i_ino);
 667}
 668
 669static const struct dentry_operations pipefs_dentry_operations = {
 670	.d_dname	= pipefs_dname,
 671};
 672
 673static struct inode * get_pipe_inode(void)
 674{
 675	struct inode *inode = new_inode_pseudo(pipe_mnt->mnt_sb);
 676	struct pipe_inode_info *pipe;
 677
 678	if (!inode)
 679		goto fail_inode;
 680
 681	inode->i_ino = get_next_ino();
 682
 683	pipe = alloc_pipe_info();
 684	if (!pipe)
 685		goto fail_iput;
 686
 687	inode->i_pipe = pipe;
 688	pipe->files = 2;
 689	pipe->readers = pipe->writers = 1;
 690	inode->i_fop = &pipefifo_fops;
 691
 692	/*
 693	 * Mark the inode dirty from the very beginning,
 694	 * that way it will never be moved to the dirty
 695	 * list because "mark_inode_dirty()" will think
 696	 * that it already _is_ on the dirty list.
 697	 */
 698	inode->i_state = I_DIRTY;
 699	inode->i_mode = S_IFIFO | S_IRUSR | S_IWUSR;
 700	inode->i_uid = current_fsuid();
 701	inode->i_gid = current_fsgid();
 702	inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
 703
 704	return inode;
 705
 706fail_iput:
 707	iput(inode);
 708
 709fail_inode:
 710	return NULL;
 711}
 712
 713int create_pipe_files(struct file **res, int flags)
 714{
 715	int err;
 716	struct inode *inode = get_pipe_inode();
 717	struct file *f;
 718	struct path path;
 719	static struct qstr name = { .name = "" };
 720
 721	if (!inode)
 722		return -ENFILE;
 723
 724	err = -ENOMEM;
 725	path.dentry = d_alloc_pseudo(pipe_mnt->mnt_sb, &name);
 726	if (!path.dentry)
 727		goto err_inode;
 728	path.mnt = mntget(pipe_mnt);
 729
 730	d_instantiate(path.dentry, inode);
 731
 732	f = alloc_file(&path, FMODE_WRITE, &pipefifo_fops);
 733	if (IS_ERR(f)) {
 734		err = PTR_ERR(f);
 735		goto err_dentry;
 736	}
 737
 738	f->f_flags = O_WRONLY | (flags & (O_NONBLOCK | O_DIRECT));
 739	f->private_data = inode->i_pipe;
 740
 741	res[0] = alloc_file(&path, FMODE_READ, &pipefifo_fops);
 742	if (IS_ERR(res[0])) {
 743		err = PTR_ERR(res[0]);
 744		goto err_file;
 745	}
 746
 747	path_get(&path);
 748	res[0]->private_data = inode->i_pipe;
 749	res[0]->f_flags = O_RDONLY | (flags & O_NONBLOCK);
 750	res[1] = f;
 751	return 0;
 752
 753err_file:
 754	put_filp(f);
 755err_dentry:
 756	free_pipe_info(inode->i_pipe);
 757	path_put(&path);
 758	return err;
 759
 760err_inode:
 761	free_pipe_info(inode->i_pipe);
 762	iput(inode);
 763	return err;
 764}
 765
 766static int __do_pipe_flags(int *fd, struct file **files, int flags)
 767{
 768	int error;
 769	int fdw, fdr;
 770
 771	if (flags & ~(O_CLOEXEC | O_NONBLOCK | O_DIRECT))
 772		return -EINVAL;
 773
 774	error = create_pipe_files(files, flags);
 775	if (error)
 776		return error;
 777
 778	error = get_unused_fd_flags(flags);
 779	if (error < 0)
 780		goto err_read_pipe;
 781	fdr = error;
 782
 783	error = get_unused_fd_flags(flags);
 784	if (error < 0)
 785		goto err_fdr;
 786	fdw = error;
 787
 788	audit_fd_pair(fdr, fdw);
 789	fd[0] = fdr;
 790	fd[1] = fdw;
 791	return 0;
 792
 793 err_fdr:
 794	put_unused_fd(fdr);
 795 err_read_pipe:
 796	fput(files[0]);
 797	fput(files[1]);
 798	return error;
 799}
 800
 801int do_pipe_flags(int *fd, int flags)
 802{
 803	struct file *files[2];
 804	int error = __do_pipe_flags(fd, files, flags);
 805	if (!error) {
 806		fd_install(fd[0], files[0]);
 807		fd_install(fd[1], files[1]);
 808	}
 809	return error;
 810}
 811
 812/*
 813 * sys_pipe() is the normal C calling standard for creating
 814 * a pipe. It's not the way Unix traditionally does this, though.
 815 */
 816SYSCALL_DEFINE2(pipe2, int __user *, fildes, int, flags)
 817{
 818	struct file *files[2];
 819	int fd[2];
 820	int error;
 821
 822	error = __do_pipe_flags(fd, files, flags);
 823	if (!error) {
 824		if (unlikely(copy_to_user(fildes, fd, sizeof(fd)))) {
 825			fput(files[0]);
 826			fput(files[1]);
 827			put_unused_fd(fd[0]);
 828			put_unused_fd(fd[1]);
 829			error = -EFAULT;
 830		} else {
 831			fd_install(fd[0], files[0]);
 832			fd_install(fd[1], files[1]);
 833		}
 834	}
 835	return error;
 836}
 837
 
 
 
 
 
 838SYSCALL_DEFINE1(pipe, int __user *, fildes)
 839{
 840	return sys_pipe2(fildes, 0);
 841}
 842
 843static int wait_for_partner(struct pipe_inode_info *pipe, unsigned int *cnt)
 844{
 845	int cur = *cnt;	
 846
 847	while (cur == *cnt) {
 848		pipe_wait(pipe);
 849		if (signal_pending(current))
 850			break;
 851	}
 852	return cur == *cnt ? -ERESTARTSYS : 0;
 853}
 854
 855static void wake_up_partner(struct pipe_inode_info *pipe)
 856{
 857	wake_up_interruptible(&pipe->wait);
 858}
 859
 860static int fifo_open(struct inode *inode, struct file *filp)
 861{
 862	struct pipe_inode_info *pipe;
 863	bool is_pipe = inode->i_sb->s_magic == PIPEFS_MAGIC;
 864	int ret;
 865
 866	filp->f_version = 0;
 867
 868	spin_lock(&inode->i_lock);
 869	if (inode->i_pipe) {
 870		pipe = inode->i_pipe;
 871		pipe->files++;
 872		spin_unlock(&inode->i_lock);
 873	} else {
 874		spin_unlock(&inode->i_lock);
 875		pipe = alloc_pipe_info();
 876		if (!pipe)
 877			return -ENOMEM;
 878		pipe->files = 1;
 879		spin_lock(&inode->i_lock);
 880		if (unlikely(inode->i_pipe)) {
 881			inode->i_pipe->files++;
 882			spin_unlock(&inode->i_lock);
 883			free_pipe_info(pipe);
 884			pipe = inode->i_pipe;
 885		} else {
 886			inode->i_pipe = pipe;
 887			spin_unlock(&inode->i_lock);
 888		}
 889	}
 890	filp->private_data = pipe;
 891	/* OK, we have a pipe and it's pinned down */
 892
 893	__pipe_lock(pipe);
 894
 895	/* We can only do regular read/write on fifos */
 896	filp->f_mode &= (FMODE_READ | FMODE_WRITE);
 897
 898	switch (filp->f_mode) {
 899	case FMODE_READ:
 900	/*
 901	 *  O_RDONLY
 902	 *  POSIX.1 says that O_NONBLOCK means return with the FIFO
 903	 *  opened, even when there is no process writing the FIFO.
 904	 */
 905		pipe->r_counter++;
 906		if (pipe->readers++ == 0)
 907			wake_up_partner(pipe);
 908
 909		if (!is_pipe && !pipe->writers) {
 910			if ((filp->f_flags & O_NONBLOCK)) {
 911				/* suppress POLLHUP until we have
 912				 * seen a writer */
 913				filp->f_version = pipe->w_counter;
 914			} else {
 915				if (wait_for_partner(pipe, &pipe->w_counter))
 916					goto err_rd;
 917			}
 918		}
 919		break;
 920	
 921	case FMODE_WRITE:
 922	/*
 923	 *  O_WRONLY
 924	 *  POSIX.1 says that O_NONBLOCK means return -1 with
 925	 *  errno=ENXIO when there is no process reading the FIFO.
 926	 */
 927		ret = -ENXIO;
 928		if (!is_pipe && (filp->f_flags & O_NONBLOCK) && !pipe->readers)
 929			goto err;
 930
 931		pipe->w_counter++;
 932		if (!pipe->writers++)
 933			wake_up_partner(pipe);
 934
 935		if (!is_pipe && !pipe->readers) {
 936			if (wait_for_partner(pipe, &pipe->r_counter))
 937				goto err_wr;
 938		}
 939		break;
 940	
 941	case FMODE_READ | FMODE_WRITE:
 942	/*
 943	 *  O_RDWR
 944	 *  POSIX.1 leaves this case "undefined" when O_NONBLOCK is set.
 945	 *  This implementation will NEVER block on a O_RDWR open, since
 946	 *  the process can at least talk to itself.
 947	 */
 948
 949		pipe->readers++;
 950		pipe->writers++;
 951		pipe->r_counter++;
 952		pipe->w_counter++;
 953		if (pipe->readers == 1 || pipe->writers == 1)
 954			wake_up_partner(pipe);
 955		break;
 956
 957	default:
 958		ret = -EINVAL;
 959		goto err;
 960	}
 961
 962	/* Ok! */
 963	__pipe_unlock(pipe);
 964	return 0;
 965
 966err_rd:
 967	if (!--pipe->readers)
 968		wake_up_interruptible(&pipe->wait);
 969	ret = -ERESTARTSYS;
 970	goto err;
 971
 972err_wr:
 973	if (!--pipe->writers)
 974		wake_up_interruptible(&pipe->wait);
 975	ret = -ERESTARTSYS;
 976	goto err;
 977
 978err:
 979	__pipe_unlock(pipe);
 980
 981	put_pipe_info(inode, pipe);
 982	return ret;
 983}
 984
 985const struct file_operations pipefifo_fops = {
 986	.open		= fifo_open,
 987	.llseek		= no_llseek,
 988	.read_iter	= pipe_read,
 989	.write_iter	= pipe_write,
 990	.poll		= pipe_poll,
 991	.unlocked_ioctl	= pipe_ioctl,
 992	.release	= pipe_release,
 993	.fasync		= pipe_fasync,
 994};
 995
 996/*
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 997 * Allocate a new array of pipe buffers and copy the info over. Returns the
 998 * pipe size if successful, or return -ERROR on error.
 999 */
1000static long pipe_set_size(struct pipe_inode_info *pipe, unsigned long nr_pages)
1001{
1002	struct pipe_buffer *bufs;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1003
1004	/*
1005	 * We can shrink the pipe, if arg >= pipe->nrbufs. Since we don't
1006	 * expect a lot of shrink+grow operations, just free and allocate
1007	 * again like we would do for growing. If the pipe currently
1008	 * contains more buffers than arg, then return busy.
1009	 */
1010	if (nr_pages < pipe->nrbufs)
1011		return -EBUSY;
 
 
1012
1013	bufs = kcalloc(nr_pages, sizeof(*bufs), GFP_KERNEL | __GFP_NOWARN);
1014	if (unlikely(!bufs))
1015		return -ENOMEM;
 
 
 
1016
1017	/*
1018	 * The pipe array wraps around, so just start the new one at zero
1019	 * and adjust the indexes.
1020	 */
1021	if (pipe->nrbufs) {
1022		unsigned int tail;
1023		unsigned int head;
1024
1025		tail = pipe->curbuf + pipe->nrbufs;
1026		if (tail < pipe->buffers)
1027			tail = 0;
1028		else
1029			tail &= (pipe->buffers - 1);
1030
1031		head = pipe->nrbufs - tail;
1032		if (head)
1033			memcpy(bufs, pipe->bufs + pipe->curbuf, head * sizeof(struct pipe_buffer));
1034		if (tail)
1035			memcpy(bufs + head, pipe->bufs, tail * sizeof(struct pipe_buffer));
1036	}
1037
1038	account_pipe_buffers(pipe, pipe->buffers, nr_pages);
1039	pipe->curbuf = 0;
1040	kfree(pipe->bufs);
1041	pipe->bufs = bufs;
1042	pipe->buffers = nr_pages;
1043	return nr_pages * PAGE_SIZE;
1044}
1045
1046/*
1047 * Currently we rely on the pipe array holding a power-of-2 number
1048 * of pages.
1049 */
1050static inline unsigned int round_pipe_size(unsigned int size)
1051{
1052	unsigned long nr_pages;
1053
1054	nr_pages = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
1055	return roundup_pow_of_two(nr_pages) << PAGE_SHIFT;
1056}
1057
1058/*
1059 * This should work even if CONFIG_PROC_FS isn't set, as proc_dointvec_minmax
1060 * will return an error.
1061 */
1062int pipe_proc_fn(struct ctl_table *table, int write, void __user *buf,
1063		 size_t *lenp, loff_t *ppos)
1064{
1065	int ret;
1066
1067	ret = proc_dointvec_minmax(table, write, buf, lenp, ppos);
1068	if (ret < 0 || !write)
1069		return ret;
1070
1071	pipe_max_size = round_pipe_size(pipe_max_size);
1072	return ret;
1073}
1074
1075/*
1076 * After the inode slimming patch, i_pipe/i_bdev/i_cdev share the same
1077 * location, so checking ->i_pipe is not enough to verify that this is a
1078 * pipe.
1079 */
1080struct pipe_inode_info *get_pipe_info(struct file *file)
1081{
1082	return file->f_op == &pipefifo_fops ? file->private_data : NULL;
1083}
1084
1085long pipe_fcntl(struct file *file, unsigned int cmd, unsigned long arg)
1086{
1087	struct pipe_inode_info *pipe;
1088	long ret;
1089
1090	pipe = get_pipe_info(file);
1091	if (!pipe)
1092		return -EBADF;
1093
1094	__pipe_lock(pipe);
1095
1096	switch (cmd) {
1097	case F_SETPIPE_SZ: {
1098		unsigned int size, nr_pages;
1099
1100		size = round_pipe_size(arg);
1101		nr_pages = size >> PAGE_SHIFT;
1102
1103		ret = -EINVAL;
1104		if (!nr_pages)
1105			goto out;
1106
1107		if (!capable(CAP_SYS_RESOURCE) && size > pipe_max_size) {
1108			ret = -EPERM;
1109			goto out;
1110		} else if ((too_many_pipe_buffers_hard(pipe->user) ||
1111			    too_many_pipe_buffers_soft(pipe->user)) &&
1112		           !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN)) {
1113			ret = -EPERM;
1114			goto out;
1115		}
1116		ret = pipe_set_size(pipe, nr_pages);
1117		break;
1118		}
1119	case F_GETPIPE_SZ:
1120		ret = pipe->buffers * PAGE_SIZE;
1121		break;
1122	default:
1123		ret = -EINVAL;
1124		break;
1125	}
1126
1127out:
1128	__pipe_unlock(pipe);
1129	return ret;
1130}
1131
1132static const struct super_operations pipefs_ops = {
1133	.destroy_inode = free_inode_nonrcu,
1134	.statfs = simple_statfs,
1135};
1136
1137/*
1138 * pipefs should _never_ be mounted by userland - too much of security hassle,
1139 * no real gain from having the whole whorehouse mounted. So we don't need
1140 * any operations on the root directory. However, we need a non-trivial
1141 * d_name - pipe: will go nicely and kill the special-casing in procfs.
1142 */
1143static struct dentry *pipefs_mount(struct file_system_type *fs_type,
1144			 int flags, const char *dev_name, void *data)
1145{
1146	return mount_pseudo(fs_type, "pipe:", &pipefs_ops,
1147			&pipefs_dentry_operations, PIPEFS_MAGIC);
1148}
1149
1150static struct file_system_type pipe_fs_type = {
1151	.name		= "pipefs",
1152	.mount		= pipefs_mount,
1153	.kill_sb	= kill_anon_super,
1154};
1155
1156static int __init init_pipe_fs(void)
1157{
1158	int err = register_filesystem(&pipe_fs_type);
1159
1160	if (!err) {
1161		pipe_mnt = kern_mount(&pipe_fs_type);
1162		if (IS_ERR(pipe_mnt)) {
1163			err = PTR_ERR(pipe_mnt);
1164			unregister_filesystem(&pipe_fs_type);
1165		}
1166	}
1167	return err;
1168}
1169
1170fs_initcall(init_pipe_fs);