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