git.oblomov.eu Git - linux-2.6/blob - fs/pipe.c

   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/mount.h>
  15 #include <linux/pipe_fs_i.h>
  16 #include <linux/uio.h>
  17 #include <linux/highmem.h>
  18 #include <linux/pagemap.h>
  19
  20 #include <asm/uaccess.h>
  21 #include <asm/ioctls.h>
  22
  23 /*
  24  * We use a start+len construction, which provides full use of the
  25  * allocated memory.
  26  * -- Florian Coosmann (FGC)
  27  *
  28  * Reads with count = 0 should always return 0.
  29  * -- Julian Bradfield 1999-06-07.
  30  *
  31  * FIFOs and Pipes now generate SIGIO for both readers and writers.
  32  * -- Jeremy Elson <jelson@circlemud.org> 2001-08-16
  33  *
  34  * pipe_read & write cleanup
  35  * -- Manfred Spraul <manfred@colorfullife.com> 2002-05-09
  36  */
  37
  38 /* Drop the inode semaphore and wait for a pipe event, atomically */
  39 void pipe_wait(struct pipe_inode_info *pipe)
  40 {
  41         DEFINE_WAIT(wait);
  42
  43         /*
  44          * Pipes are system-local resources, so sleeping on them
  45          * is considered a noninteractive wait:
  46          */
  47         prepare_to_wait(&pipe->wait, &wait,
  48                         TASK_INTERRUPTIBLE | TASK_NONINTERACTIVE);
  49         if (pipe->inode)
  50                 mutex_unlock(&pipe->inode->i_mutex);
  51         schedule();
  52         finish_wait(&pipe->wait, &wait);
  53         if (pipe->inode)
  54                 mutex_lock(&pipe->inode->i_mutex);
  55 }
  56
  57 static int
  58 pipe_iov_copy_from_user(void *to, struct iovec *iov, unsigned long len)
  59 {
  60         unsigned long copy;
  61
  62         while (len > 0) {
  63                 while (!iov->iov_len)
  64                         iov++;
  65                 copy = min_t(unsigned long, len, iov->iov_len);
  66
  67                 if (copy_from_user(to, iov->iov_base, copy))
  68                         return -EFAULT;
  69                 to += copy;
  70                 len -= copy;
  71                 iov->iov_base += copy;
  72                 iov->iov_len -= copy;
  73         }
  74         return 0;
  75 }
  76
  77 static int
  78 pipe_iov_copy_to_user(struct iovec *iov, const void *from, unsigned long len)
  79 {
  80         unsigned long copy;
  81
  82         while (len > 0) {
  83                 while (!iov->iov_len)
  84                         iov++;
  85                 copy = min_t(unsigned long, len, iov->iov_len);
  86
  87                 if (copy_to_user(iov->iov_base, from, copy))
  88                         return -EFAULT;
  89                 from += copy;
  90                 len -= copy;
  91                 iov->iov_base += copy;
  92                 iov->iov_len -= copy;
  93         }
  94         return 0;
  95 }
  96
  97 static void anon_pipe_buf_release(struct pipe_inode_info *pipe,
  98                                   struct pipe_buffer *buf)
  99 {
 100         struct page *page = buf->page;
 101
 102         buf->flags &= ~PIPE_BUF_FLAG_STOLEN;
 103
 104         /*
 105          * If nobody else uses this page, and we don't already have a
 106          * temporary page, let's keep track of it as a one-deep
 107          * allocation cache. (Otherwise just release our reference to it)
 108          */
 109         if (page_count(page) == 1 && !pipe->tmp_page)
 110                 pipe->tmp_page = page;
 111         else
 112                 page_cache_release(page);
 113 }
 114
 115 static void * anon_pipe_buf_map(struct file *file, struct pipe_inode_info *pipe,
 116                                 struct pipe_buffer *buf)
 117 {
 118         return kmap(buf->page);
 119 }
 120
 121 static void anon_pipe_buf_unmap(struct pipe_inode_info *pipe,
 122                                 struct pipe_buffer *buf)
 123 {
 124         kunmap(buf->page);
 125 }
 126
 127 static int anon_pipe_buf_steal(struct pipe_inode_info *pipe,
 128                                struct pipe_buffer *buf)
 129 {
 130         buf->flags |= PIPE_BUF_FLAG_STOLEN;
 131         return 0;
 132 }
 133
 134 static struct pipe_buf_operations anon_pipe_buf_ops = {
 135         .can_merge = 1,
 136         .map = anon_pipe_buf_map,
 137         .unmap = anon_pipe_buf_unmap,
 138         .release = anon_pipe_buf_release,
 139         .steal = anon_pipe_buf_steal,
 140 };
 141
 142 static ssize_t
 143 pipe_readv(struct file *filp, const struct iovec *_iov,
 144            unsigned long nr_segs, loff_t *ppos)
 145 {
 146         struct inode *inode = filp->f_dentry->d_inode;
 147         struct pipe_inode_info *pipe;
 148         int do_wakeup;
 149         ssize_t ret;
 150         struct iovec *iov = (struct iovec *)_iov;
 151         size_t total_len;
 152
 153         total_len = iov_length(iov, nr_segs);
 154         /* Null read succeeds. */
 155         if (unlikely(total_len == 0))
 156                 return 0;
 157
 158         do_wakeup = 0;
 159         ret = 0;
 160         mutex_lock(&inode->i_mutex);
 161         pipe = inode->i_pipe;
 162         for (;;) {
 163                 int bufs = pipe->nrbufs;
 164                 if (bufs) {
 165                         int curbuf = pipe->curbuf;
 166                         struct pipe_buffer *buf = pipe->bufs + curbuf;
 167                         struct pipe_buf_operations *ops = buf->ops;
 168                         void *addr;
 169                         size_t chars = buf->len;
 170                         int error;
 171
 172                         if (chars > total_len)
 173                                 chars = total_len;
 174
 175                         addr = ops->map(filp, pipe, buf);
 176                         if (IS_ERR(addr)) {
 177                                 if (!ret)
 178                                         ret = PTR_ERR(addr);
 179                                 break;
 180                         }
 181                         error = pipe_iov_copy_to_user(iov, addr + buf->offset, chars);
 182                         ops->unmap(pipe, buf);
 183                         if (unlikely(error)) {
 184                                 if (!ret)
 185                                         ret = -EFAULT;
 186                                 break;
 187                         }
 188                         ret += chars;
 189                         buf->offset += chars;
 190                         buf->len -= chars;
 191                         if (!buf->len) {
 192                                 buf->ops = NULL;
 193                                 ops->release(pipe, buf);
 194                                 curbuf = (curbuf + 1) & (PIPE_BUFFERS-1);
 195                                 pipe->curbuf = curbuf;
 196                                 pipe->nrbufs = --bufs;
 197                                 do_wakeup = 1;
 198                         }
 199                         total_len -= chars;
 200                         if (!total_len)
 201                                 break;  /* common path: read succeeded */
 202                 }
 203                 if (bufs)       /* More to do? */
 204                         continue;
 205                 if (!pipe->writers)
 206                         break;
 207                 if (!pipe->waiting_writers) {
 208                         /* syscall merging: Usually we must not sleep
 209                          * if O_NONBLOCK is set, or if we got some data.
 210                          * But if a writer sleeps in kernel space, then
 211                          * we can wait for that data without violating POSIX.
 212                          */
 213                         if (ret)
 214                                 break;
 215                         if (filp->f_flags & O_NONBLOCK) {
 216                                 ret = -EAGAIN;
 217                                 break;
 218                         }
 219                 }
 220                 if (signal_pending(current)) {
 221                         if (!ret)
 222                                 ret = -ERESTARTSYS;
 223                         break;
 224                 }
 225                 if (do_wakeup) {
 226                         wake_up_interruptible_sync(&pipe->wait);
 227                         kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
 228                 }
 229                 pipe_wait(pipe);
 230         }
 231         mutex_unlock(&inode->i_mutex);
 232
 233         /* Signal writers asynchronously that there is more room. */
 234         if (do_wakeup) {
 235                 wake_up_interruptible(&pipe->wait);
 236                 kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
 237         }
 238         if (ret > 0)
 239                 file_accessed(filp);
 240         return ret;
 241 }
 242
 243 static ssize_t
 244 pipe_read(struct file *filp, char __user *buf, size_t count, loff_t *ppos)
 245 {
 246         struct iovec iov = { .iov_base = buf, .iov_len = count };
 247
 248         return pipe_readv(filp, &iov, 1, ppos);
 249 }
 250
 251 static ssize_t
 252 pipe_writev(struct file *filp, const struct iovec *_iov,
 253             unsigned long nr_segs, loff_t *ppos)
 254 {
 255         struct inode *inode = filp->f_dentry->d_inode;
 256         struct pipe_inode_info *pipe;
 257         ssize_t ret;
 258         int do_wakeup;
 259         struct iovec *iov = (struct iovec *)_iov;
 260         size_t total_len;
 261         ssize_t chars;
 262
 263         total_len = iov_length(iov, nr_segs);
 264         /* Null write succeeds. */
 265         if (unlikely(total_len == 0))
 266                 return 0;
 267
 268         do_wakeup = 0;
 269         ret = 0;
 270         mutex_lock(&inode->i_mutex);
 271         pipe = inode->i_pipe;
 272
 273         if (!pipe->readers) {
 274                 send_sig(SIGPIPE, current, 0);
 275                 ret = -EPIPE;
 276                 goto out;
 277         }
 278
 279         /* We try to merge small writes */
 280         chars = total_len & (PAGE_SIZE-1); /* size of the last buffer */
 281         if (pipe->nrbufs && chars != 0) {
 282                 int lastbuf = (pipe->curbuf + pipe->nrbufs - 1) &
 283                                                         (PIPE_BUFFERS-1);
 284                 struct pipe_buffer *buf = pipe->bufs + lastbuf;
 285                 struct pipe_buf_operations *ops = buf->ops;
 286                 int offset = buf->offset + buf->len;
 287
 288                 if (ops->can_merge && offset + chars <= PAGE_SIZE) {
 289                         void *addr;
 290                         int error;
 291
 292                         addr = ops->map(filp, pipe, buf);
 293                         if (IS_ERR(addr)) {
 294                                 error = PTR_ERR(addr);
 295                                 goto out;
 296                         }
 297                         error = pipe_iov_copy_from_user(offset + addr, iov,
 298                                                         chars);
 299                         ops->unmap(pipe, buf);
 300                         ret = error;
 301                         do_wakeup = 1;
 302                         if (error)
 303                                 goto out;
 304                         buf->len += chars;
 305                         total_len -= chars;
 306                         ret = chars;
 307                         if (!total_len)
 308                                 goto out;
 309                 }
 310         }
 311
 312         for (;;) {
 313                 int bufs;
 314
 315                 if (!pipe->readers) {
 316                         send_sig(SIGPIPE, current, 0);
 317                         if (!ret)
 318                                 ret = -EPIPE;
 319                         break;
 320                 }
 321                 bufs = pipe->nrbufs;
 322                 if (bufs < PIPE_BUFFERS) {
 323                         int newbuf = (pipe->curbuf + bufs) & (PIPE_BUFFERS-1);
 324                         struct pipe_buffer *buf = pipe->bufs + newbuf;
 325                         struct page *page = pipe->tmp_page;
 326                         int error;
 327
 328                         if (!page) {
 329                                 page = alloc_page(GFP_HIGHUSER);
 330                                 if (unlikely(!page)) {
 331                                         ret = ret ? : -ENOMEM;
 332                                         break;
 333                                 }
 334                                 pipe->tmp_page = page;
 335                         }
 336                         /* Always wake up, even if the copy fails. Otherwise
 337                          * we lock up (O_NONBLOCK-)readers that sleep due to
 338                          * syscall merging.
 339                          * FIXME! Is this really true?
 340                          */
 341                         do_wakeup = 1;
 342                         chars = PAGE_SIZE;
 343                         if (chars > total_len)
 344                                 chars = total_len;
 345
 346                         error = pipe_iov_copy_from_user(kmap(page), iov, chars);
 347                         kunmap(page);
 348                         if (unlikely(error)) {
 349                                 if (!ret)
 350                                         ret = -EFAULT;
 351                                 break;
 352                         }
 353                         ret += chars;
 354
 355                         /* Insert it into the buffer array */
 356                         buf->page = page;
 357                         buf->ops = &anon_pipe_buf_ops;
 358                         buf->offset = 0;
 359                         buf->len = chars;
 360                         pipe->nrbufs = ++bufs;
 361                         pipe->tmp_page = NULL;
 362
 363                         total_len -= chars;
 364                         if (!total_len)
 365                                 break;
 366                 }
 367                 if (bufs < PIPE_BUFFERS)
 368                         continue;
 369                 if (filp->f_flags & O_NONBLOCK) {
 370                         if (!ret)
 371                                 ret = -EAGAIN;
 372                         break;
 373                 }
 374                 if (signal_pending(current)) {
 375                         if (!ret)
 376                                 ret = -ERESTARTSYS;
 377                         break;
 378                 }
 379                 if (do_wakeup) {
 380                         wake_up_interruptible_sync(&pipe->wait);
 381                         kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
 382                         do_wakeup = 0;
 383                 }
 384                 pipe->waiting_writers++;
 385                 pipe_wait(pipe);
 386                 pipe->waiting_writers--;
 387         }
 388 out:
 389         mutex_unlock(&inode->i_mutex);
 390         if (do_wakeup) {
 391                 wake_up_interruptible(&pipe->wait);
 392                 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
 393         }
 394         if (ret > 0)
 395                 file_update_time(filp);
 396         return ret;
 397 }
 398
 399 static ssize_t
 400 pipe_write(struct file *filp, const char __user *buf,
 401            size_t count, loff_t *ppos)
 402 {
 403         struct iovec iov = { .iov_base = (void __user *)buf, .iov_len = count };
 404
 405         return pipe_writev(filp, &iov, 1, ppos);
 406 }
 407
 408 static ssize_t
 409 bad_pipe_r(struct file *filp, char __user *buf, size_t count, loff_t *ppos)
 410 {
 411         return -EBADF;
 412 }
 413
 414 static ssize_t
 415 bad_pipe_w(struct file *filp, const char __user *buf, size_t count,
 416            loff_t *ppos)
 417 {
 418         return -EBADF;
 419 }
 420
 421 static int
 422 pipe_ioctl(struct inode *pino, struct file *filp,
 423            unsigned int cmd, unsigned long arg)
 424 {
 425         struct inode *inode = filp->f_dentry->d_inode;
 426         struct pipe_inode_info *pipe;
 427         int count, buf, nrbufs;
 428
 429         switch (cmd) {
 430                 case FIONREAD:
 431                         mutex_lock(&inode->i_mutex);
 432                         pipe = inode->i_pipe;
 433                         count = 0;
 434                         buf = pipe->curbuf;
 435                         nrbufs = pipe->nrbufs;
 436                         while (--nrbufs >= 0) {
 437                                 count += pipe->bufs[buf].len;
 438                                 buf = (buf+1) & (PIPE_BUFFERS-1);
 439                         }
 440                         mutex_unlock(&inode->i_mutex);
 441
 442                         return put_user(count, (int __user *)arg);
 443                 default:
 444                         return -EINVAL;
 445         }
 446 }
 447
 448 /* No kernel lock held - fine */
 449 static unsigned int
 450 pipe_poll(struct file *filp, poll_table *wait)
 451 {
 452         unsigned int mask;
 453         struct inode *inode = filp->f_dentry->d_inode;
 454         struct pipe_inode_info *pipe = inode->i_pipe;
 455         int nrbufs;
 456
 457         poll_wait(filp, &pipe->wait, wait);
 458
 459         /* Reading only -- no need for acquiring the semaphore.  */
 460         nrbufs = pipe->nrbufs;
 461         mask = 0;
 462         if (filp->f_mode & FMODE_READ) {
 463                 mask = (nrbufs > 0) ? POLLIN | POLLRDNORM : 0;
 464                 if (!pipe->writers && filp->f_version != pipe->w_counter)
 465                         mask |= POLLHUP;
 466         }
 467
 468         if (filp->f_mode & FMODE_WRITE) {
 469                 mask |= (nrbufs < PIPE_BUFFERS) ? POLLOUT | POLLWRNORM : 0;
 470                 /*
 471                  * Most Unices do not set POLLERR for FIFOs but on Linux they
 472                  * behave exactly like pipes for poll().
 473                  */
 474                 if (!pipe->readers)
 475                         mask |= POLLERR;
 476         }
 477
 478         return mask;
 479 }
 480
 481 static int
 482 pipe_release(struct inode *inode, int decr, int decw)
 483 {
 484         struct pipe_inode_info *pipe;
 485
 486         mutex_lock(&inode->i_mutex);
 487         pipe = inode->i_pipe;
 488         pipe->readers -= decr;
 489         pipe->writers -= decw;
 490
 491         if (!pipe->readers && !pipe->writers) {
 492                 free_pipe_info(inode);
 493         } else {
 494                 wake_up_interruptible(&pipe->wait);
 495                 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
 496                 kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
 497         }
 498         mutex_unlock(&inode->i_mutex);
 499
 500         return 0;
 501 }
 502
 503 static int
 504 pipe_read_fasync(int fd, struct file *filp, int on)
 505 {
 506         struct inode *inode = filp->f_dentry->d_inode;
 507         int retval;
 508
 509         mutex_lock(&inode->i_mutex);
 510         retval = fasync_helper(fd, filp, on, &inode->i_pipe->fasync_readers);
 511         mutex_unlock(&inode->i_mutex);
 512
 513         if (retval < 0)
 514                 return retval;
 515
 516         return 0;
 517 }
 518
 519
 520 static int
 521 pipe_write_fasync(int fd, struct file *filp, int on)
 522 {
 523         struct inode *inode = filp->f_dentry->d_inode;
 524         int retval;
 525
 526         mutex_lock(&inode->i_mutex);
 527         retval = fasync_helper(fd, filp, on, &inode->i_pipe->fasync_writers);
 528         mutex_unlock(&inode->i_mutex);
 529
 530         if (retval < 0)
 531                 return retval;
 532
 533         return 0;
 534 }
 535
 536
 537 static int
 538 pipe_rdwr_fasync(int fd, struct file *filp, int on)
 539 {
 540         struct inode *inode = filp->f_dentry->d_inode;
 541         struct pipe_inode_info *pipe = inode->i_pipe;
 542         int retval;
 543
 544         mutex_lock(&inode->i_mutex);
 545
 546         retval = fasync_helper(fd, filp, on, &pipe->fasync_readers);
 547
 548         if (retval >= 0)
 549                 retval = fasync_helper(fd, filp, on, &pipe->fasync_writers);
 550
 551         mutex_unlock(&inode->i_mutex);
 552
 553         if (retval < 0)
 554                 return retval;
 555
 556         return 0;
 557 }
 558
 559
 560 static int
 561 pipe_read_release(struct inode *inode, struct file *filp)
 562 {
 563         pipe_read_fasync(-1, filp, 0);
 564         return pipe_release(inode, 1, 0);
 565 }
 566
 567 static int
 568 pipe_write_release(struct inode *inode, struct file *filp)
 569 {
 570         pipe_write_fasync(-1, filp, 0);
 571         return pipe_release(inode, 0, 1);
 572 }
 573
 574 static int
 575 pipe_rdwr_release(struct inode *inode, struct file *filp)
 576 {
 577         int decr, decw;
 578
 579         pipe_rdwr_fasync(-1, filp, 0);
 580         decr = (filp->f_mode & FMODE_READ) != 0;
 581         decw = (filp->f_mode & FMODE_WRITE) != 0;
 582         return pipe_release(inode, decr, decw);
 583 }
 584
 585 static int
 586 pipe_read_open(struct inode *inode, struct file *filp)
 587 {
 588         /* We could have perhaps used atomic_t, but this and friends
 589            below are the only places.  So it doesn't seem worthwhile.  */
 590         mutex_lock(&inode->i_mutex);
 591         inode->i_pipe->readers++;
 592         mutex_unlock(&inode->i_mutex);
 593
 594         return 0;
 595 }
 596
 597 static int
 598 pipe_write_open(struct inode *inode, struct file *filp)
 599 {
 600         mutex_lock(&inode->i_mutex);
 601         inode->i_pipe->writers++;
 602         mutex_unlock(&inode->i_mutex);
 603
 604         return 0;
 605 }
 606
 607 static int
 608 pipe_rdwr_open(struct inode *inode, struct file *filp)
 609 {
 610         mutex_lock(&inode->i_mutex);
 611         if (filp->f_mode & FMODE_READ)
 612                 inode->i_pipe->readers++;
 613         if (filp->f_mode & FMODE_WRITE)
 614                 inode->i_pipe->writers++;
 615         mutex_unlock(&inode->i_mutex);
 616
 617         return 0;
 618 }
 619
 620 /*
 621  * The file_operations structs are not static because they
 622  * are also used in linux/fs/fifo.c to do operations on FIFOs.
 623  */
 624 const struct file_operations read_fifo_fops = {
 625         .llseek         = no_llseek,
 626         .read           = pipe_read,
 627         .readv          = pipe_readv,
 628         .write          = bad_pipe_w,
 629         .poll           = pipe_poll,
 630         .ioctl          = pipe_ioctl,
 631         .open           = pipe_read_open,
 632         .release        = pipe_read_release,
 633         .fasync         = pipe_read_fasync,
 634 };
 635
 636 const struct file_operations write_fifo_fops = {
 637         .llseek         = no_llseek,
 638         .read           = bad_pipe_r,
 639         .write          = pipe_write,
 640         .writev         = pipe_writev,
 641         .poll           = pipe_poll,
 642         .ioctl          = pipe_ioctl,
 643         .open           = pipe_write_open,
 644         .release        = pipe_write_release,
 645         .fasync         = pipe_write_fasync,
 646 };
 647
 648 const struct file_operations rdwr_fifo_fops = {
 649         .llseek         = no_llseek,
 650         .read           = pipe_read,
 651         .readv          = pipe_readv,
 652         .write          = pipe_write,
 653         .writev         = pipe_writev,
 654         .poll           = pipe_poll,
 655         .ioctl          = pipe_ioctl,
 656         .open           = pipe_rdwr_open,
 657         .release        = pipe_rdwr_release,
 658         .fasync         = pipe_rdwr_fasync,
 659 };
 660
 661 static struct file_operations read_pipe_fops = {
 662         .llseek         = no_llseek,
 663         .read           = pipe_read,
 664         .readv          = pipe_readv,
 665         .write          = bad_pipe_w,
 666         .poll           = pipe_poll,
 667         .ioctl          = pipe_ioctl,
 668         .open           = pipe_read_open,
 669         .release        = pipe_read_release,
 670         .fasync         = pipe_read_fasync,
 671 };
 672
 673 static struct file_operations write_pipe_fops = {
 674         .llseek         = no_llseek,
 675         .read           = bad_pipe_r,
 676         .write          = pipe_write,
 677         .writev         = pipe_writev,
 678         .poll           = pipe_poll,
 679         .ioctl          = pipe_ioctl,
 680         .open           = pipe_write_open,
 681         .release        = pipe_write_release,
 682         .fasync         = pipe_write_fasync,
 683 };
 684
 685 static struct file_operations rdwr_pipe_fops = {
 686         .llseek         = no_llseek,
 687         .read           = pipe_read,
 688         .readv          = pipe_readv,
 689         .write          = pipe_write,
 690         .writev         = pipe_writev,
 691         .poll           = pipe_poll,
 692         .ioctl          = pipe_ioctl,
 693         .open           = pipe_rdwr_open,
 694         .release        = pipe_rdwr_release,
 695         .fasync         = pipe_rdwr_fasync,
 696 };
 697
 698 struct pipe_inode_info * alloc_pipe_info(struct inode *inode)
 699 {
 700         struct pipe_inode_info *pipe;
 701
 702         pipe = kzalloc(sizeof(struct pipe_inode_info), GFP_KERNEL);
 703         if (pipe) {
 704                 init_waitqueue_head(&pipe->wait);
 705                 pipe->r_counter = pipe->w_counter = 1;
 706                 pipe->inode = inode;
 707         }
 708
 709         return pipe;
 710 }
 711
 712 void __free_pipe_info(struct pipe_inode_info *pipe)
 713 {
 714         int i;
 715
 716         for (i = 0; i < PIPE_BUFFERS; i++) {
 717                 struct pipe_buffer *buf = pipe->bufs + i;
 718                 if (buf->ops)
 719                         buf->ops->release(pipe, buf);
 720         }
 721         if (pipe->tmp_page)
 722                 __free_page(pipe->tmp_page);
 723         kfree(pipe);
 724 }
 725
 726 void free_pipe_info(struct inode *inode)
 727 {
 728         __free_pipe_info(inode->i_pipe);
 729         inode->i_pipe = NULL;
 730 }
 731
 732 static struct vfsmount *pipe_mnt __read_mostly;
 733 static int pipefs_delete_dentry(struct dentry *dentry)
 734 {
 735         return 1;
 736 }
 737
 738 static struct dentry_operations pipefs_dentry_operations = {
 739         .d_delete       = pipefs_delete_dentry,
 740 };
 741
 742 static struct inode * get_pipe_inode(void)
 743 {
 744         struct inode *inode = new_inode(pipe_mnt->mnt_sb);
 745         struct pipe_inode_info *pipe;
 746
 747         if (!inode)
 748                 goto fail_inode;
 749
 750         pipe = alloc_pipe_info(inode);
 751         if (!pipe)
 752                 goto fail_iput;
 753         inode->i_pipe = pipe;
 754
 755         pipe->readers = pipe->writers = 1;
 756         inode->i_fop = &rdwr_pipe_fops;
 757
 758         /*
 759          * Mark the inode dirty from the very beginning,
 760          * that way it will never be moved to the dirty
 761          * list because "mark_inode_dirty()" will think
 762          * that it already _is_ on the dirty list.
 763          */
 764         inode->i_state = I_DIRTY;
 765         inode->i_mode = S_IFIFO | S_IRUSR | S_IWUSR;
 766         inode->i_uid = current->fsuid;
 767         inode->i_gid = current->fsgid;
 768         inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
 769         inode->i_blksize = PAGE_SIZE;
 770
 771         return inode;
 772
 773 fail_iput:
 774         iput(inode);
 775
 776 fail_inode:
 777         return NULL;
 778 }
 779
 780 int do_pipe(int *fd)
 781 {
 782         struct qstr this;
 783         char name[32];
 784         struct dentry *dentry;
 785         struct inode * inode;
 786         struct file *f1, *f2;
 787         int error;
 788         int i, j;
 789
 790         error = -ENFILE;
 791         f1 = get_empty_filp();
 792         if (!f1)
 793                 goto no_files;
 794
 795         f2 = get_empty_filp();
 796         if (!f2)
 797                 goto close_f1;
 798
 799         inode = get_pipe_inode();
 800         if (!inode)
 801                 goto close_f12;
 802
 803         error = get_unused_fd();
 804         if (error < 0)
 805                 goto close_f12_inode;
 806         i = error;
 807
 808         error = get_unused_fd();
 809         if (error < 0)
 810                 goto close_f12_inode_i;
 811         j = error;
 812
 813         error = -ENOMEM;
 814         sprintf(name, "[%lu]", inode->i_ino);
 815         this.name = name;
 816         this.len = strlen(name);
 817         this.hash = inode->i_ino; /* will go */
 818         dentry = d_alloc(pipe_mnt->mnt_sb->s_root, &this);
 819         if (!dentry)
 820                 goto close_f12_inode_i_j;
 821
 822         dentry->d_op = &pipefs_dentry_operations;
 823         d_add(dentry, inode);
 824         f1->f_vfsmnt = f2->f_vfsmnt = mntget(mntget(pipe_mnt));
 825         f1->f_dentry = f2->f_dentry = dget(dentry);
 826         f1->f_mapping = f2->f_mapping = inode->i_mapping;
 827
 828         /* read file */
 829         f1->f_pos = f2->f_pos = 0;
 830         f1->f_flags = O_RDONLY;
 831         f1->f_op = &read_pipe_fops;
 832         f1->f_mode = FMODE_READ;
 833         f1->f_version = 0;
 834
 835         /* write file */
 836         f2->f_flags = O_WRONLY;
 837         f2->f_op = &write_pipe_fops;
 838         f2->f_mode = FMODE_WRITE;
 839         f2->f_version = 0;
 840
 841         fd_install(i, f1);
 842         fd_install(j, f2);
 843         fd[0] = i;
 844         fd[1] = j;
 845
 846         return 0;
 847
 848 close_f12_inode_i_j:
 849         put_unused_fd(j);
 850 close_f12_inode_i:
 851         put_unused_fd(i);
 852 close_f12_inode:
 853         free_pipe_info(inode);
 854         iput(inode);
 855 close_f12:
 856         put_filp(f2);
 857 close_f1:
 858         put_filp(f1);
 859 no_files:
 860         return error;
 861 }
 862
 863 /*
 864  * pipefs should _never_ be mounted by userland - too much of security hassle,
 865  * no real gain from having the whole whorehouse mounted. So we don't need
 866  * any operations on the root directory. However, we need a non-trivial
 867  * d_name - pipe: will go nicely and kill the special-casing in procfs.
 868  */
 869
 870 static struct super_block *
 871 pipefs_get_sb(struct file_system_type *fs_type, int flags,
 872               const char *dev_name, void *data)
 873 {
 874         return get_sb_pseudo(fs_type, "pipe:", NULL, PIPEFS_MAGIC);
 875 }
 876
 877 static struct file_system_type pipe_fs_type = {
 878         .name           = "pipefs",
 879         .get_sb         = pipefs_get_sb,
 880         .kill_sb        = kill_anon_super,
 881 };
 882
 883 static int __init init_pipe_fs(void)
 884 {
 885         int err = register_filesystem(&pipe_fs_type);
 886
 887         if (!err) {
 888                 pipe_mnt = kern_mount(&pipe_fs_type);
 889                 if (IS_ERR(pipe_mnt)) {
 890                         err = PTR_ERR(pipe_mnt);
 891                         unregister_filesystem(&pipe_fs_type);
 892                 }
 893         }
 894         return err;
 895 }
 896
 897 static void __exit exit_pipe_fs(void)
 898 {
 899         unregister_filesystem(&pipe_fs_type);
 900         mntput(pipe_mnt);
 901 }
 902
 903 fs_initcall(init_pipe_fs);
 904 module_exit(exit_pipe_fs);