pipe.c 24.1 KB
Newer Older
L
Linus Torvalds 已提交
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17
/*
 *  linux/fs/pipe.c
 *
 *  Copyright (C) 1991, 1992, 1999  Linus Torvalds
 */

#include <linux/mm.h>
#include <linux/file.h>
#include <linux/poll.h>
#include <linux/slab.h>
#include <linux/module.h>
#include <linux/init.h>
#include <linux/fs.h>
#include <linux/mount.h>
#include <linux/pipe_fs_i.h>
#include <linux/uio.h>
#include <linux/highmem.h>
18
#include <linux/pagemap.h>
A
Al Viro 已提交
19
#include <linux/audit.h>
20
#include <linux/syscalls.h>
L
Linus Torvalds 已提交
21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40

#include <asm/uaccess.h>
#include <asm/ioctls.h>

/*
 * We use a start+len construction, which provides full use of the 
 * allocated memory.
 * -- Florian Coosmann (FGC)
 * 
 * Reads with count = 0 should always return 0.
 * -- Julian Bradfield 1999-06-07.
 *
 * FIFOs and Pipes now generate SIGIO for both readers and writers.
 * -- Jeremy Elson <jelson@circlemud.org> 2001-08-16
 *
 * pipe_read & write cleanup
 * -- Manfred Spraul <manfred@colorfullife.com> 2002-05-09
 */

/* Drop the inode semaphore and wait for a pipe event, atomically */
41
void pipe_wait(struct pipe_inode_info *pipe)
L
Linus Torvalds 已提交
42 43 44
{
	DEFINE_WAIT(wait);

I
Ingo Molnar 已提交
45 46 47 48
	/*
	 * Pipes are system-local resources, so sleeping on them
	 * is considered a noninteractive wait:
	 */
49
	prepare_to_wait(&pipe->wait, &wait, TASK_INTERRUPTIBLE);
50 51
	if (pipe->inode)
		mutex_unlock(&pipe->inode->i_mutex);
L
Linus Torvalds 已提交
52
	schedule();
53 54 55
	finish_wait(&pipe->wait, &wait);
	if (pipe->inode)
		mutex_lock(&pipe->inode->i_mutex);
L
Linus Torvalds 已提交
56 57
}

58
static int
59 60
pipe_iov_copy_from_user(void *to, struct iovec *iov, unsigned long len,
			int atomic)
L
Linus Torvalds 已提交
61 62 63 64 65 66 67 68
{
	unsigned long copy;

	while (len > 0) {
		while (!iov->iov_len)
			iov++;
		copy = min_t(unsigned long, len, iov->iov_len);

69 70 71 72 73 74 75
		if (atomic) {
			if (__copy_from_user_inatomic(to, iov->iov_base, copy))
				return -EFAULT;
		} else {
			if (copy_from_user(to, iov->iov_base, copy))
				return -EFAULT;
		}
L
Linus Torvalds 已提交
76 77 78 79 80 81 82 83
		to += copy;
		len -= copy;
		iov->iov_base += copy;
		iov->iov_len -= copy;
	}
	return 0;
}

84
static int
85 86
pipe_iov_copy_to_user(struct iovec *iov, const void *from, unsigned long len,
		      int atomic)
L
Linus Torvalds 已提交
87 88 89 90 91 92 93 94
{
	unsigned long copy;

	while (len > 0) {
		while (!iov->iov_len)
			iov++;
		copy = min_t(unsigned long, len, iov->iov_len);

95 96 97 98 99 100 101
		if (atomic) {
			if (__copy_to_user_inatomic(iov->iov_base, from, copy))
				return -EFAULT;
		} else {
			if (copy_to_user(iov->iov_base, from, copy))
				return -EFAULT;
		}
L
Linus Torvalds 已提交
102 103 104 105 106 107 108 109
		from += copy;
		len -= copy;
		iov->iov_base += copy;
		iov->iov_len -= copy;
	}
	return 0;
}

110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150
/*
 * Attempt to pre-fault in the user memory, so we can use atomic copies.
 * Returns the number of bytes not faulted in.
 */
static int iov_fault_in_pages_write(struct iovec *iov, unsigned long len)
{
	while (!iov->iov_len)
		iov++;

	while (len > 0) {
		unsigned long this_len;

		this_len = min_t(unsigned long, len, iov->iov_len);
		if (fault_in_pages_writeable(iov->iov_base, this_len))
			break;

		len -= this_len;
		iov++;
	}

	return len;
}

/*
 * Pre-fault in the user memory, so we can use atomic copies.
 */
static void iov_fault_in_pages_read(struct iovec *iov, unsigned long len)
{
	while (!iov->iov_len)
		iov++;

	while (len > 0) {
		unsigned long this_len;

		this_len = min_t(unsigned long, len, iov->iov_len);
		fault_in_pages_readable(iov->iov_base, this_len);
		len -= this_len;
		iov++;
	}
}

151 152
static void anon_pipe_buf_release(struct pipe_inode_info *pipe,
				  struct pipe_buffer *buf)
L
Linus Torvalds 已提交
153 154 155
{
	struct page *page = buf->page;

156 157 158
	/*
	 * If nobody else uses this page, and we don't already have a
	 * temporary page, let's keep track of it as a one-deep
159
	 * allocation cache. (Otherwise just release our reference to it)
160
	 */
161
	if (page_count(page) == 1 && !pipe->tmp_page)
162
		pipe->tmp_page = page;
163 164
	else
		page_cache_release(page);
L
Linus Torvalds 已提交
165 166
}

167 168 169 170 171 172 173 174
/**
 * generic_pipe_buf_map - virtually map a pipe buffer
 * @pipe:	the pipe that the buffer belongs to
 * @buf:	the buffer that should be mapped
 * @atomic:	whether to use an atomic map
 *
 * Description:
 *	This function returns a kernel virtual address mapping for the
175
 *	pipe_buffer passed in @buf. If @atomic is set, an atomic map is provided
176 177 178 179 180
 *	and the caller has to be careful not to fault before calling
 *	the unmap function.
 *
 *	Note that this function occupies KM_USER0 if @atomic != 0.
 */
181
void *generic_pipe_buf_map(struct pipe_inode_info *pipe,
182
			   struct pipe_buffer *buf, int atomic)
L
Linus Torvalds 已提交
183
{
184 185 186 187 188
	if (atomic) {
		buf->flags |= PIPE_BUF_FLAG_ATOMIC;
		return kmap_atomic(buf->page, KM_USER0);
	}

L
Linus Torvalds 已提交
189 190 191
	return kmap(buf->page);
}

192 193 194 195 196 197 198 199 200
/**
 * generic_pipe_buf_unmap - unmap a previously mapped pipe buffer
 * @pipe:	the pipe that the buffer belongs to
 * @buf:	the buffer that should be unmapped
 * @map_data:	the data that the mapping function returned
 *
 * Description:
 *	This function undoes the mapping that ->map() provided.
 */
201
void generic_pipe_buf_unmap(struct pipe_inode_info *pipe,
202
			    struct pipe_buffer *buf, void *map_data)
L
Linus Torvalds 已提交
203
{
204 205 206 207 208
	if (buf->flags & PIPE_BUF_FLAG_ATOMIC) {
		buf->flags &= ~PIPE_BUF_FLAG_ATOMIC;
		kunmap_atomic(map_data, KM_USER0);
	} else
		kunmap(buf->page);
L
Linus Torvalds 已提交
209 210
}

211
/**
212
 * generic_pipe_buf_steal - attempt to take ownership of a &pipe_buffer
213 214 215 216
 * @pipe:	the pipe that the buffer belongs to
 * @buf:	the buffer to attempt to steal
 *
 * Description:
217
 *	This function attempts to steal the &struct page attached to
218 219
 *	@buf. If successful, this function returns 0 and returns with
 *	the page locked. The caller may then reuse the page for whatever
220
 *	he wishes; the typical use is insertion into a different file
221 222
 *	page cache.
 */
223 224
int generic_pipe_buf_steal(struct pipe_inode_info *pipe,
			   struct pipe_buffer *buf)
225
{
226 227
	struct page *page = buf->page;

228 229 230 231 232
	/*
	 * A reference of one is golden, that means that the owner of this
	 * page is the only one holding a reference to it. lock the page
	 * and return OK.
	 */
233 234 235 236 237 238
	if (page_count(page) == 1) {
		lock_page(page);
		return 0;
	}

	return 1;
239 240
}

241
/**
242
 * generic_pipe_buf_get - get a reference to a &struct pipe_buffer
243 244 245 246 247 248 249 250 251
 * @pipe:	the pipe that the buffer belongs to
 * @buf:	the buffer to get a reference to
 *
 * Description:
 *	This function grabs an extra reference to @buf. It's used in
 *	in the tee() system call, when we duplicate the buffers in one
 *	pipe into another.
 */
void generic_pipe_buf_get(struct pipe_inode_info *pipe, struct pipe_buffer *buf)
252 253 254 255
{
	page_cache_get(buf->page);
}

256 257
/**
 * generic_pipe_buf_confirm - verify contents of the pipe buffer
R
Randy Dunlap 已提交
258
 * @info:	the pipe that the buffer belongs to
259 260 261 262 263 264
 * @buf:	the buffer to confirm
 *
 * Description:
 *	This function does nothing, because the generic pipe code uses
 *	pages that are always good when inserted into the pipe.
 */
265 266
int generic_pipe_buf_confirm(struct pipe_inode_info *info,
			     struct pipe_buffer *buf)
267 268 269 270
{
	return 0;
}

271
static const struct pipe_buf_operations anon_pipe_buf_ops = {
L
Linus Torvalds 已提交
272
	.can_merge = 1,
273 274
	.map = generic_pipe_buf_map,
	.unmap = generic_pipe_buf_unmap,
275
	.confirm = generic_pipe_buf_confirm,
L
Linus Torvalds 已提交
276
	.release = anon_pipe_buf_release,
277
	.steal = generic_pipe_buf_steal,
278
	.get = generic_pipe_buf_get,
L
Linus Torvalds 已提交
279 280 281
};

static ssize_t
282 283
pipe_read(struct kiocb *iocb, const struct iovec *_iov,
	   unsigned long nr_segs, loff_t pos)
L
Linus Torvalds 已提交
284
{
285
	struct file *filp = iocb->ki_filp;
286
	struct inode *inode = filp->f_path.dentry->d_inode;
287
	struct pipe_inode_info *pipe;
L
Linus Torvalds 已提交
288 289 290 291 292 293 294 295 296 297 298 299
	int do_wakeup;
	ssize_t ret;
	struct iovec *iov = (struct iovec *)_iov;
	size_t total_len;

	total_len = iov_length(iov, nr_segs);
	/* Null read succeeds. */
	if (unlikely(total_len == 0))
		return 0;

	do_wakeup = 0;
	ret = 0;
300
	mutex_lock(&inode->i_mutex);
301
	pipe = inode->i_pipe;
L
Linus Torvalds 已提交
302
	for (;;) {
303
		int bufs = pipe->nrbufs;
L
Linus Torvalds 已提交
304
		if (bufs) {
305 306
			int curbuf = pipe->curbuf;
			struct pipe_buffer *buf = pipe->bufs + curbuf;
307
			const struct pipe_buf_operations *ops = buf->ops;
L
Linus Torvalds 已提交
308 309
			void *addr;
			size_t chars = buf->len;
310
			int error, atomic;
L
Linus Torvalds 已提交
311 312 313 314

			if (chars > total_len)
				chars = total_len;

315
			error = ops->confirm(pipe, buf);
316
			if (error) {
317
				if (!ret)
318
					error = ret;
319 320
				break;
			}
321

322 323 324 325 326
			atomic = !iov_fault_in_pages_write(iov, chars);
redo:
			addr = ops->map(pipe, buf, atomic);
			error = pipe_iov_copy_to_user(iov, addr + buf->offset, chars, atomic);
			ops->unmap(pipe, buf, addr);
L
Linus Torvalds 已提交
327
			if (unlikely(error)) {
328 329 330 331 332 333 334
				/*
				 * Just retry with the slow path if we failed.
				 */
				if (atomic) {
					atomic = 0;
					goto redo;
				}
335
				if (!ret)
336
					ret = error;
L
Linus Torvalds 已提交
337 338 339 340 341 342 343
				break;
			}
			ret += chars;
			buf->offset += chars;
			buf->len -= chars;
			if (!buf->len) {
				buf->ops = NULL;
344
				ops->release(pipe, buf);
L
Linus Torvalds 已提交
345
				curbuf = (curbuf + 1) & (PIPE_BUFFERS-1);
346 347
				pipe->curbuf = curbuf;
				pipe->nrbufs = --bufs;
L
Linus Torvalds 已提交
348 349 350 351 352 353 354 355
				do_wakeup = 1;
			}
			total_len -= chars;
			if (!total_len)
				break;	/* common path: read succeeded */
		}
		if (bufs)	/* More to do? */
			continue;
356
		if (!pipe->writers)
L
Linus Torvalds 已提交
357
			break;
358
		if (!pipe->waiting_writers) {
L
Linus Torvalds 已提交
359 360 361 362 363 364 365 366 367 368 369 370 371
			/* syscall merging: Usually we must not sleep
			 * if O_NONBLOCK is set, or if we got some data.
			 * But if a writer sleeps in kernel space, then
			 * we can wait for that data without violating POSIX.
			 */
			if (ret)
				break;
			if (filp->f_flags & O_NONBLOCK) {
				ret = -EAGAIN;
				break;
			}
		}
		if (signal_pending(current)) {
372 373
			if (!ret)
				ret = -ERESTARTSYS;
L
Linus Torvalds 已提交
374 375 376
			break;
		}
		if (do_wakeup) {
377 378
			wake_up_interruptible_sync(&pipe->wait);
 			kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
L
Linus Torvalds 已提交
379
		}
380
		pipe_wait(pipe);
L
Linus Torvalds 已提交
381
	}
382
	mutex_unlock(&inode->i_mutex);
383 384

	/* Signal writers asynchronously that there is more room. */
L
Linus Torvalds 已提交
385
	if (do_wakeup) {
I
Ingo Molnar 已提交
386
		wake_up_interruptible_sync(&pipe->wait);
387
		kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
L
Linus Torvalds 已提交
388 389 390 391 392 393 394
	}
	if (ret > 0)
		file_accessed(filp);
	return ret;
}

static ssize_t
395 396
pipe_write(struct kiocb *iocb, const struct iovec *_iov,
	    unsigned long nr_segs, loff_t ppos)
L
Linus Torvalds 已提交
397
{
398
	struct file *filp = iocb->ki_filp;
399
	struct inode *inode = filp->f_path.dentry->d_inode;
400
	struct pipe_inode_info *pipe;
L
Linus Torvalds 已提交
401 402 403 404 405 406 407 408 409 410 411 412 413
	ssize_t ret;
	int do_wakeup;
	struct iovec *iov = (struct iovec *)_iov;
	size_t total_len;
	ssize_t chars;

	total_len = iov_length(iov, nr_segs);
	/* Null write succeeds. */
	if (unlikely(total_len == 0))
		return 0;

	do_wakeup = 0;
	ret = 0;
414
	mutex_lock(&inode->i_mutex);
415
	pipe = inode->i_pipe;
L
Linus Torvalds 已提交
416

417
	if (!pipe->readers) {
L
Linus Torvalds 已提交
418 419 420 421 422 423 424
		send_sig(SIGPIPE, current, 0);
		ret = -EPIPE;
		goto out;
	}

	/* We try to merge small writes */
	chars = total_len & (PAGE_SIZE-1); /* size of the last buffer */
425
	if (pipe->nrbufs && chars != 0) {
426 427
		int lastbuf = (pipe->curbuf + pipe->nrbufs - 1) &
							(PIPE_BUFFERS-1);
428
		struct pipe_buffer *buf = pipe->bufs + lastbuf;
429
		const struct pipe_buf_operations *ops = buf->ops;
L
Linus Torvalds 已提交
430
		int offset = buf->offset + buf->len;
431

L
Linus Torvalds 已提交
432
		if (ops->can_merge && offset + chars <= PAGE_SIZE) {
433
			int error, atomic = 1;
434 435
			void *addr;

436
			error = ops->confirm(pipe, buf);
437
			if (error)
438
				goto out;
439

440 441 442
			iov_fault_in_pages_read(iov, chars);
redo1:
			addr = ops->map(pipe, buf, atomic);
443
			error = pipe_iov_copy_from_user(offset + addr, iov,
444 445
							chars, atomic);
			ops->unmap(pipe, buf, addr);
L
Linus Torvalds 已提交
446 447
			ret = error;
			do_wakeup = 1;
448 449 450 451 452
			if (error) {
				if (atomic) {
					atomic = 0;
					goto redo1;
				}
L
Linus Torvalds 已提交
453
				goto out;
454
			}
L
Linus Torvalds 已提交
455 456 457 458 459 460 461 462 463 464
			buf->len += chars;
			total_len -= chars;
			ret = chars;
			if (!total_len)
				goto out;
		}
	}

	for (;;) {
		int bufs;
465

466
		if (!pipe->readers) {
L
Linus Torvalds 已提交
467
			send_sig(SIGPIPE, current, 0);
468 469
			if (!ret)
				ret = -EPIPE;
L
Linus Torvalds 已提交
470 471
			break;
		}
472
		bufs = pipe->nrbufs;
L
Linus Torvalds 已提交
473
		if (bufs < PIPE_BUFFERS) {
474 475 476
			int newbuf = (pipe->curbuf + bufs) & (PIPE_BUFFERS-1);
			struct pipe_buffer *buf = pipe->bufs + newbuf;
			struct page *page = pipe->tmp_page;
477 478
			char *src;
			int error, atomic = 1;
L
Linus Torvalds 已提交
479 480 481 482 483 484 485

			if (!page) {
				page = alloc_page(GFP_HIGHUSER);
				if (unlikely(!page)) {
					ret = ret ? : -ENOMEM;
					break;
				}
486
				pipe->tmp_page = page;
L
Linus Torvalds 已提交
487
			}
488
			/* Always wake up, even if the copy fails. Otherwise
L
Linus Torvalds 已提交
489 490 491 492 493 494 495 496 497
			 * we lock up (O_NONBLOCK-)readers that sleep due to
			 * syscall merging.
			 * FIXME! Is this really true?
			 */
			do_wakeup = 1;
			chars = PAGE_SIZE;
			if (chars > total_len)
				chars = total_len;

498 499 500 501 502 503 504 505 506 507 508 509 510 511
			iov_fault_in_pages_read(iov, chars);
redo2:
			if (atomic)
				src = kmap_atomic(page, KM_USER0);
			else
				src = kmap(page);

			error = pipe_iov_copy_from_user(src, iov, chars,
							atomic);
			if (atomic)
				kunmap_atomic(src, KM_USER0);
			else
				kunmap(page);

L
Linus Torvalds 已提交
512
			if (unlikely(error)) {
513 514 515 516
				if (atomic) {
					atomic = 0;
					goto redo2;
				}
517
				if (!ret)
518
					ret = error;
L
Linus Torvalds 已提交
519 520 521 522 523 524 525 526 527
				break;
			}
			ret += chars;

			/* Insert it into the buffer array */
			buf->page = page;
			buf->ops = &anon_pipe_buf_ops;
			buf->offset = 0;
			buf->len = chars;
528 529
			pipe->nrbufs = ++bufs;
			pipe->tmp_page = NULL;
L
Linus Torvalds 已提交
530 531 532 533 534 535 536 537

			total_len -= chars;
			if (!total_len)
				break;
		}
		if (bufs < PIPE_BUFFERS)
			continue;
		if (filp->f_flags & O_NONBLOCK) {
538 539
			if (!ret)
				ret = -EAGAIN;
L
Linus Torvalds 已提交
540 541 542
			break;
		}
		if (signal_pending(current)) {
543 544
			if (!ret)
				ret = -ERESTARTSYS;
L
Linus Torvalds 已提交
545 546 547
			break;
		}
		if (do_wakeup) {
548 549
			wake_up_interruptible_sync(&pipe->wait);
			kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
L
Linus Torvalds 已提交
550 551
			do_wakeup = 0;
		}
552 553 554
		pipe->waiting_writers++;
		pipe_wait(pipe);
		pipe->waiting_writers--;
L
Linus Torvalds 已提交
555 556
	}
out:
557
	mutex_unlock(&inode->i_mutex);
L
Linus Torvalds 已提交
558
	if (do_wakeup) {
I
Ingo Molnar 已提交
559
		wake_up_interruptible_sync(&pipe->wait);
560
		kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
L
Linus Torvalds 已提交
561 562
	}
	if (ret > 0)
563
		file_update_time(filp);
L
Linus Torvalds 已提交
564 565 566 567 568 569 570 571 572 573
	return ret;
}

static ssize_t
bad_pipe_r(struct file *filp, char __user *buf, size_t count, loff_t *ppos)
{
	return -EBADF;
}

static ssize_t
574 575
bad_pipe_w(struct file *filp, const char __user *buf, size_t count,
	   loff_t *ppos)
L
Linus Torvalds 已提交
576 577 578 579
{
	return -EBADF;
}

580
static long pipe_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
L
Linus Torvalds 已提交
581
{
582
	struct inode *inode = filp->f_path.dentry->d_inode;
583
	struct pipe_inode_info *pipe;
L
Linus Torvalds 已提交
584 585 586 587
	int count, buf, nrbufs;

	switch (cmd) {
		case FIONREAD:
588
			mutex_lock(&inode->i_mutex);
589
			pipe = inode->i_pipe;
L
Linus Torvalds 已提交
590
			count = 0;
591 592
			buf = pipe->curbuf;
			nrbufs = pipe->nrbufs;
L
Linus Torvalds 已提交
593
			while (--nrbufs >= 0) {
594
				count += pipe->bufs[buf].len;
L
Linus Torvalds 已提交
595 596
				buf = (buf+1) & (PIPE_BUFFERS-1);
			}
597
			mutex_unlock(&inode->i_mutex);
598

L
Linus Torvalds 已提交
599 600 601 602 603 604 605 606 607 608 609
			return put_user(count, (int __user *)arg);
		default:
			return -EINVAL;
	}
}

/* No kernel lock held - fine */
static unsigned int
pipe_poll(struct file *filp, poll_table *wait)
{
	unsigned int mask;
610
	struct inode *inode = filp->f_path.dentry->d_inode;
611
	struct pipe_inode_info *pipe = inode->i_pipe;
L
Linus Torvalds 已提交
612 613
	int nrbufs;

614
	poll_wait(filp, &pipe->wait, wait);
L
Linus Torvalds 已提交
615 616

	/* Reading only -- no need for acquiring the semaphore.  */
617
	nrbufs = pipe->nrbufs;
L
Linus Torvalds 已提交
618 619 620
	mask = 0;
	if (filp->f_mode & FMODE_READ) {
		mask = (nrbufs > 0) ? POLLIN | POLLRDNORM : 0;
621
		if (!pipe->writers && filp->f_version != pipe->w_counter)
L
Linus Torvalds 已提交
622 623 624 625 626
			mask |= POLLHUP;
	}

	if (filp->f_mode & FMODE_WRITE) {
		mask |= (nrbufs < PIPE_BUFFERS) ? POLLOUT | POLLWRNORM : 0;
627 628 629 630
		/*
		 * Most Unices do not set POLLERR for FIFOs but on Linux they
		 * behave exactly like pipes for poll().
		 */
631
		if (!pipe->readers)
L
Linus Torvalds 已提交
632 633 634 635 636 637 638 639 640
			mask |= POLLERR;
	}

	return mask;
}

static int
pipe_release(struct inode *inode, int decr, int decw)
{
641 642
	struct pipe_inode_info *pipe;

643
	mutex_lock(&inode->i_mutex);
644 645 646
	pipe = inode->i_pipe;
	pipe->readers -= decr;
	pipe->writers -= decw;
647

648
	if (!pipe->readers && !pipe->writers) {
L
Linus Torvalds 已提交
649 650
		free_pipe_info(inode);
	} else {
I
Ingo Molnar 已提交
651
		wake_up_interruptible_sync(&pipe->wait);
652 653
		kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
		kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
L
Linus Torvalds 已提交
654
	}
655
	mutex_unlock(&inode->i_mutex);
L
Linus Torvalds 已提交
656 657 658 659 660 661 662

	return 0;
}

static int
pipe_read_fasync(int fd, struct file *filp, int on)
{
663
	struct inode *inode = filp->f_path.dentry->d_inode;
L
Linus Torvalds 已提交
664 665
	int retval;

666 667 668
	mutex_lock(&inode->i_mutex);
	retval = fasync_helper(fd, filp, on, &inode->i_pipe->fasync_readers);
	mutex_unlock(&inode->i_mutex);
L
Linus Torvalds 已提交
669 670 671 672 673 674 675 676 677 678 679

	if (retval < 0)
		return retval;

	return 0;
}


static int
pipe_write_fasync(int fd, struct file *filp, int on)
{
680
	struct inode *inode = filp->f_path.dentry->d_inode;
L
Linus Torvalds 已提交
681 682
	int retval;

683 684 685
	mutex_lock(&inode->i_mutex);
	retval = fasync_helper(fd, filp, on, &inode->i_pipe->fasync_writers);
	mutex_unlock(&inode->i_mutex);
L
Linus Torvalds 已提交
686 687 688 689 690 691 692 693 694 695 696

	if (retval < 0)
		return retval;

	return 0;
}


static int
pipe_rdwr_fasync(int fd, struct file *filp, int on)
{
697
	struct inode *inode = filp->f_path.dentry->d_inode;
698
	struct pipe_inode_info *pipe = inode->i_pipe;
L
Linus Torvalds 已提交
699 700
	int retval;

701
	mutex_lock(&inode->i_mutex);
702
	retval = fasync_helper(fd, filp, on, &pipe->fasync_readers);
703
	if (retval >= 0) {
704
		retval = fasync_helper(fd, filp, on, &pipe->fasync_writers);
705 706 707
		if (retval < 0) /* this can happen only if on == T */
			fasync_helper(-1, filp, 0, &pipe->fasync_readers);
	}
708
	mutex_unlock(&inode->i_mutex);
L
Linus Torvalds 已提交
709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743

	if (retval < 0)
		return retval;

	return 0;
}


static int
pipe_read_release(struct inode *inode, struct file *filp)
{
	return pipe_release(inode, 1, 0);
}

static int
pipe_write_release(struct inode *inode, struct file *filp)
{
	return pipe_release(inode, 0, 1);
}

static int
pipe_rdwr_release(struct inode *inode, struct file *filp)
{
	int decr, decw;

	decr = (filp->f_mode & FMODE_READ) != 0;
	decw = (filp->f_mode & FMODE_WRITE) != 0;
	return pipe_release(inode, decr, decw);
}

static int
pipe_read_open(struct inode *inode, struct file *filp)
{
	/* We could have perhaps used atomic_t, but this and friends
	   below are the only places.  So it doesn't seem worthwhile.  */
744 745 746
	mutex_lock(&inode->i_mutex);
	inode->i_pipe->readers++;
	mutex_unlock(&inode->i_mutex);
L
Linus Torvalds 已提交
747 748 749 750 751 752 753

	return 0;
}

static int
pipe_write_open(struct inode *inode, struct file *filp)
{
754 755 756
	mutex_lock(&inode->i_mutex);
	inode->i_pipe->writers++;
	mutex_unlock(&inode->i_mutex);
L
Linus Torvalds 已提交
757 758 759 760 761 762 763

	return 0;
}

static int
pipe_rdwr_open(struct inode *inode, struct file *filp)
{
764
	mutex_lock(&inode->i_mutex);
L
Linus Torvalds 已提交
765
	if (filp->f_mode & FMODE_READ)
766
		inode->i_pipe->readers++;
L
Linus Torvalds 已提交
767
	if (filp->f_mode & FMODE_WRITE)
768 769
		inode->i_pipe->writers++;
	mutex_unlock(&inode->i_mutex);
L
Linus Torvalds 已提交
770 771 772 773 774 775 776

	return 0;
}

/*
 * The file_operations structs are not static because they
 * are also used in linux/fs/fifo.c to do operations on FIFOs.
777 778
 *
 * Pipes reuse fifos' file_operations structs.
L
Linus Torvalds 已提交
779
 */
780
const struct file_operations read_pipefifo_fops = {
L
Linus Torvalds 已提交
781
	.llseek		= no_llseek,
782 783
	.read		= do_sync_read,
	.aio_read	= pipe_read,
L
Linus Torvalds 已提交
784 785
	.write		= bad_pipe_w,
	.poll		= pipe_poll,
786
	.unlocked_ioctl	= pipe_ioctl,
L
Linus Torvalds 已提交
787 788 789 790 791
	.open		= pipe_read_open,
	.release	= pipe_read_release,
	.fasync		= pipe_read_fasync,
};

792
const struct file_operations write_pipefifo_fops = {
L
Linus Torvalds 已提交
793 794
	.llseek		= no_llseek,
	.read		= bad_pipe_r,
795 796
	.write		= do_sync_write,
	.aio_write	= pipe_write,
L
Linus Torvalds 已提交
797
	.poll		= pipe_poll,
798
	.unlocked_ioctl	= pipe_ioctl,
L
Linus Torvalds 已提交
799 800 801 802 803
	.open		= pipe_write_open,
	.release	= pipe_write_release,
	.fasync		= pipe_write_fasync,
};

804
const struct file_operations rdwr_pipefifo_fops = {
L
Linus Torvalds 已提交
805
	.llseek		= no_llseek,
806 807 808 809
	.read		= do_sync_read,
	.aio_read	= pipe_read,
	.write		= do_sync_write,
	.aio_write	= pipe_write,
L
Linus Torvalds 已提交
810
	.poll		= pipe_poll,
811
	.unlocked_ioctl	= pipe_ioctl,
L
Linus Torvalds 已提交
812 813 814 815 816
	.open		= pipe_rdwr_open,
	.release	= pipe_rdwr_release,
	.fasync		= pipe_rdwr_fasync,
};

817 818
struct pipe_inode_info * alloc_pipe_info(struct inode *inode)
{
819
	struct pipe_inode_info *pipe;
820

821 822 823 824 825
	pipe = kzalloc(sizeof(struct pipe_inode_info), GFP_KERNEL);
	if (pipe) {
		init_waitqueue_head(&pipe->wait);
		pipe->r_counter = pipe->w_counter = 1;
		pipe->inode = inode;
826 827
	}

828
	return pipe;
829 830
}

831
void __free_pipe_info(struct pipe_inode_info *pipe)
L
Linus Torvalds 已提交
832 833 834 835
{
	int i;

	for (i = 0; i < PIPE_BUFFERS; i++) {
836
		struct pipe_buffer *buf = pipe->bufs + i;
L
Linus Torvalds 已提交
837
		if (buf->ops)
838
			buf->ops->release(pipe, buf);
L
Linus Torvalds 已提交
839
	}
840 841 842
	if (pipe->tmp_page)
		__free_page(pipe->tmp_page);
	kfree(pipe);
L
Linus Torvalds 已提交
843 844
}

845 846 847 848 849 850
void free_pipe_info(struct inode *inode)
{
	__free_pipe_info(inode->i_pipe);
	inode->i_pipe = NULL;
}

851
static struct vfsmount *pipe_mnt __read_mostly;
L
Linus Torvalds 已提交
852 853
static int pipefs_delete_dentry(struct dentry *dentry)
{
854 855 856 857 858 859 860 861
	/*
	 * At creation time, we pretended this dentry was hashed
	 * (by clearing DCACHE_UNHASHED bit in d_flags)
	 * At delete time, we restore the truth : not hashed.
	 * (so that dput() can proceed correctly)
	 */
	dentry->d_flags |= DCACHE_UNHASHED;
	return 0;
L
Linus Torvalds 已提交
862
}
863

864 865 866 867 868 869 870 871 872
/*
 * pipefs_dname() is called from d_path().
 */
static char *pipefs_dname(struct dentry *dentry, char *buffer, int buflen)
{
	return dynamic_dname(dentry, buffer, buflen, "pipe:[%lu]",
				dentry->d_inode->i_ino);
}

L
Linus Torvalds 已提交
873 874
static struct dentry_operations pipefs_dentry_operations = {
	.d_delete	= pipefs_delete_dentry,
875
	.d_dname	= pipefs_dname,
L
Linus Torvalds 已提交
876 877 878 879 880
};

static struct inode * get_pipe_inode(void)
{
	struct inode *inode = new_inode(pipe_mnt->mnt_sb);
881
	struct pipe_inode_info *pipe;
L
Linus Torvalds 已提交
882 883 884 885

	if (!inode)
		goto fail_inode;

886 887
	pipe = alloc_pipe_info(inode);
	if (!pipe)
L
Linus Torvalds 已提交
888
		goto fail_iput;
889
	inode->i_pipe = pipe;
890

891
	pipe->readers = pipe->writers = 1;
892
	inode->i_fop = &rdwr_pipefifo_fops;
L
Linus Torvalds 已提交
893 894 895 896 897 898 899 900 901

	/*
	 * Mark the inode dirty from the very beginning,
	 * that way it will never be moved to the dirty
	 * list because "mark_inode_dirty()" will think
	 * that it already _is_ on the dirty list.
	 */
	inode->i_state = I_DIRTY;
	inode->i_mode = S_IFIFO | S_IRUSR | S_IWUSR;
902 903
	inode->i_uid = current_fsuid();
	inode->i_gid = current_fsgid();
L
Linus Torvalds 已提交
904
	inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
905

L
Linus Torvalds 已提交
906 907 908 909
	return inode;

fail_iput:
	iput(inode);
910

L
Linus Torvalds 已提交
911 912 913 914
fail_inode:
	return NULL;
}

915
struct file *create_write_pipe(int flags)
L
Linus Torvalds 已提交
916
{
A
Andi Kleen 已提交
917 918 919
	int err;
	struct inode *inode;
	struct file *f;
L
Linus Torvalds 已提交
920
	struct dentry *dentry;
921
	struct qstr name = { .name = "" };
L
Linus Torvalds 已提交
922

A
Andi Kleen 已提交
923
	err = -ENFILE;
L
Linus Torvalds 已提交
924 925
	inode = get_pipe_inode();
	if (!inode)
D
Dave Hansen 已提交
926
		goto err;
L
Linus Torvalds 已提交
927

A
Andi Kleen 已提交
928
	err = -ENOMEM;
929
	dentry = d_alloc(pipe_mnt->mnt_sb->s_root, &name);
L
Linus Torvalds 已提交
930
	if (!dentry)
A
Andi Kleen 已提交
931
		goto err_inode;
932

L
Linus Torvalds 已提交
933
	dentry->d_op = &pipefs_dentry_operations;
934 935 936 937 938 939 940
	/*
	 * We dont want to publish this dentry into global dentry hash table.
	 * We pretend dentry is already hashed, by unsetting DCACHE_UNHASHED
	 * This permits a working /proc/$pid/fd/XXX on pipes
	 */
	dentry->d_flags &= ~DCACHE_UNHASHED;
	d_instantiate(dentry, inode);
D
Dave Hansen 已提交
941 942

	err = -ENFILE;
943
	f = alloc_file(pipe_mnt, dentry, FMODE_WRITE, &write_pipefifo_fops);
D
Dave Hansen 已提交
944 945
	if (!f)
		goto err_dentry;
A
Andi Kleen 已提交
946
	f->f_mapping = inode->i_mapping;
947

948
	f->f_flags = O_WRONLY | (flags & O_NONBLOCK);
A
Andi Kleen 已提交
949 950 951
	f->f_version = 0;

	return f;
L
Linus Torvalds 已提交
952

D
Dave Hansen 已提交
953
 err_dentry:
954
	free_pipe_info(inode);
D
Dave Hansen 已提交
955
	dput(dentry);
956 957
	return ERR_PTR(err);

A
Andi Kleen 已提交
958
 err_inode:
L
Linus Torvalds 已提交
959 960
	free_pipe_info(inode);
	iput(inode);
D
Dave Hansen 已提交
961
 err:
A
Andi Kleen 已提交
962 963 964 965 966
	return ERR_PTR(err);
}

void free_write_pipe(struct file *f)
{
967
	free_pipe_info(f->f_dentry->d_inode);
968
	path_put(&f->f_path);
A
Andi Kleen 已提交
969 970 971
	put_filp(f);
}

972
struct file *create_read_pipe(struct file *wrf, int flags)
A
Andi Kleen 已提交
973 974 975 976 977 978
{
	struct file *f = get_empty_filp();
	if (!f)
		return ERR_PTR(-ENFILE);

	/* Grab pipe from the writer */
979 980
	f->f_path = wrf->f_path;
	path_get(&wrf->f_path);
981
	f->f_mapping = wrf->f_path.dentry->d_inode->i_mapping;
A
Andi Kleen 已提交
982 983

	f->f_pos = 0;
984
	f->f_flags = O_RDONLY | (flags & O_NONBLOCK);
985
	f->f_op = &read_pipefifo_fops;
A
Andi Kleen 已提交
986 987 988 989 990 991
	f->f_mode = FMODE_READ;
	f->f_version = 0;

	return f;
}

U
Ulrich Drepper 已提交
992
int do_pipe_flags(int *fd, int flags)
A
Andi Kleen 已提交
993 994 995 996 997
{
	struct file *fw, *fr;
	int error;
	int fdw, fdr;

998
	if (flags & ~(O_CLOEXEC | O_NONBLOCK))
U
Ulrich Drepper 已提交
999 1000
		return -EINVAL;

1001
	fw = create_write_pipe(flags);
A
Andi Kleen 已提交
1002 1003
	if (IS_ERR(fw))
		return PTR_ERR(fw);
1004
	fr = create_read_pipe(fw, flags);
A
Andi Kleen 已提交
1005 1006 1007 1008
	error = PTR_ERR(fr);
	if (IS_ERR(fr))
		goto err_write_pipe;

U
Ulrich Drepper 已提交
1009
	error = get_unused_fd_flags(flags);
A
Andi Kleen 已提交
1010 1011 1012 1013
	if (error < 0)
		goto err_read_pipe;
	fdr = error;

U
Ulrich Drepper 已提交
1014
	error = get_unused_fd_flags(flags);
A
Andi Kleen 已提交
1015 1016 1017 1018
	if (error < 0)
		goto err_fdr;
	fdw = error;

A
Al Viro 已提交
1019
	audit_fd_pair(fdr, fdw);
A
Andi Kleen 已提交
1020 1021 1022 1023 1024 1025 1026 1027 1028 1029
	fd_install(fdr, fr);
	fd_install(fdw, fw);
	fd[0] = fdr;
	fd[1] = fdw;

	return 0;

 err_fdr:
	put_unused_fd(fdr);
 err_read_pipe:
1030
	path_put(&fr->f_path);
A
Andi Kleen 已提交
1031 1032 1033 1034
	put_filp(fr);
 err_write_pipe:
	free_write_pipe(fw);
	return error;
L
Linus Torvalds 已提交
1035 1036
}

1037 1038 1039 1040
/*
 * sys_pipe() is the normal C calling standard for creating
 * a pipe. It's not the way Unix traditionally does this, though.
 */
1041
SYSCALL_DEFINE2(pipe2, int __user *, fildes, int, flags)
1042 1043 1044 1045
{
	int fd[2];
	int error;

U
Ulrich Drepper 已提交
1046
	error = do_pipe_flags(fd, flags);
1047
	if (!error) {
1048 1049 1050
		if (copy_to_user(fildes, fd, sizeof(fd))) {
			sys_close(fd[0]);
			sys_close(fd[1]);
1051
			error = -EFAULT;
1052
		}
1053 1054 1055 1056
	}
	return error;
}

1057
SYSCALL_DEFINE1(pipe, int __user *, fildes)
U
Ulrich Drepper 已提交
1058 1059 1060 1061
{
	return sys_pipe2(fildes, 0);
}

L
Linus Torvalds 已提交
1062 1063 1064 1065 1066 1067
/*
 * pipefs should _never_ be mounted by userland - too much of security hassle,
 * no real gain from having the whole whorehouse mounted. So we don't need
 * any operations on the root directory. However, we need a non-trivial
 * d_name - pipe: will go nicely and kill the special-casing in procfs.
 */
1068 1069 1070
static int pipefs_get_sb(struct file_system_type *fs_type,
			 int flags, const char *dev_name, void *data,
			 struct vfsmount *mnt)
L
Linus Torvalds 已提交
1071
{
1072
	return get_sb_pseudo(fs_type, "pipe:", NULL, PIPEFS_MAGIC, mnt);
L
Linus Torvalds 已提交
1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083
}

static struct file_system_type pipe_fs_type = {
	.name		= "pipefs",
	.get_sb		= pipefs_get_sb,
	.kill_sb	= kill_anon_super,
};

static int __init init_pipe_fs(void)
{
	int err = register_filesystem(&pipe_fs_type);
1084

L
Linus Torvalds 已提交
1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102
	if (!err) {
		pipe_mnt = kern_mount(&pipe_fs_type);
		if (IS_ERR(pipe_mnt)) {
			err = PTR_ERR(pipe_mnt);
			unregister_filesystem(&pipe_fs_type);
		}
	}
	return err;
}

static void __exit exit_pipe_fs(void)
{
	unregister_filesystem(&pipe_fs_type);
	mntput(pipe_mnt);
}

fs_initcall(init_pipe_fs);
module_exit(exit_pipe_fs);