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 <asm/uaccess.h>
  17 #include <asm/ioctls.h>
  18
  19 /*
  20  * We use a start+len construction, which provides full use of the
  21  * allocated memory.
  22  * -- Florian Coosmann (FGC)
  23  *
  24  * Reads with count = 0 should always return 0.
  25  * -- Julian Bradfield 1999-06-07.
  26  *
  27  * FIFOs and Pipes now generate SIGIO for both readers and writers.
  28  * -- Jeremy Elson <jelson@circlemud.org> 2001-08-16
  29  *
  30  * pipe_read & write cleanup
  31  * -- Manfred Spraul <manfred@colorfullife.com> 2002-05-09
  32  */
  33
  34 /* Drop the inode semaphore and wait for a pipe event, atomically */
  35 void pipe_wait(struct inode * inode)
  36 {
  37         DEFINE_WAIT(wait);
  38
  39         prepare_to_wait(PIPE_WAIT(*inode), &wait, TASK_INTERRUPTIBLE);
  40         up(PIPE_SEM(*inode));
  41         schedule();
  42         finish_wait(PIPE_WAIT(*inode), &wait);
  43         down(PIPE_SEM(*inode));
  44 }
  45
  46 static ssize_t
  47 pipe_read(struct file *filp, char *buf, size_t count, loff_t *ppos)
  48 {
  49         struct inode *inode = filp->f_dentry->d_inode;
  50         int do_wakeup;
  51         ssize_t ret;
  52
  53         /* pread is not allowed on pipes. */
  54         if (unlikely(ppos != &filp->f_pos))
  55                 return -ESPIPE;
  56
  57         /* Null read succeeds. */
  58         if (unlikely(count == 0))
  59                 return 0;
  60
  61         do_wakeup = 0;
  62         ret = 0;
  63         down(PIPE_SEM(*inode));
  64         for (;;) {
  65                 int size = PIPE_LEN(*inode);
  66                 if (size) {
  67                         char *pipebuf = PIPE_BASE(*inode) + PIPE_START(*inode);
  68                         ssize_t chars = PIPE_MAX_RCHUNK(*inode);
  69
  70                         if (chars > count)
  71                                 chars = count;
  72                         if (chars > size)
  73                                 chars = size;
  74
  75                         if (copy_to_user(buf, pipebuf, chars)) {
  76                                 if (!ret) ret = -EFAULT;
  77                                 break;
  78                         }
  79                         ret += chars;
  80
  81                         PIPE_START(*inode) += chars;
  82                         PIPE_START(*inode) &= (PIPE_SIZE - 1);
  83                         PIPE_LEN(*inode) -= chars;
  84                         count -= chars;
  85                         buf += chars;
  86                         do_wakeup = 1;
  87                 }
  88                 if (!count)
  89                         break;  /* common path: read succeeded */
  90                 if (PIPE_LEN(*inode)) /* test for cyclic buffers */
  91                         continue;
  92                 if (!PIPE_WRITERS(*inode))
  93                         break;
  94                 if (!PIPE_WAITING_WRITERS(*inode)) {
  95                         /* syscall merging: Usually we must not sleep
  96                          * if O_NONBLOCK is set, or if we got some data.
  97                          * But if a writer sleeps in kernel space, then
  98                          * we can wait for that data without violating POSIX.
  99                          */
 100                         if (ret)
 101                                 break;
 102                         if (filp->f_flags & O_NONBLOCK) {
 103                                 ret = -EAGAIN;
 104                                 break;
 105                         }
 106                 }
 107                 if (signal_pending(current)) {
 108                         if (!ret) ret = -ERESTARTSYS;
 109                         break;
 110                 }
 111                 if (do_wakeup) {
 112                         wake_up_interruptible_sync(PIPE_WAIT(*inode));
 113                         kill_fasync(PIPE_FASYNC_WRITERS(*inode), SIGIO, POLL_OUT);
 114                 }
 115                 pipe_wait(inode);
 116         }
 117         up(PIPE_SEM(*inode));
 118         /* Signal writers asynchronously that there is more room.  */
 119         if (do_wakeup) {
 120                 wake_up_interruptible(PIPE_WAIT(*inode));
 121                 kill_fasync(PIPE_FASYNC_WRITERS(*inode), SIGIO, POLL_OUT);
 122         }
 123         if (ret > 0)
 124                 UPDATE_ATIME(inode);
 125         return ret;
 126 }
 127
 128 static ssize_t
 129 pipe_write(struct file *filp, const char *buf, size_t count, loff_t *ppos)
 130 {
 131         struct inode *inode = filp->f_dentry->d_inode;
 132         ssize_t ret;
 133         size_t min;
 134         int do_wakeup;
 135
 136         /* pwrite is not allowed on pipes. */
 137         if (unlikely(ppos != &filp->f_pos))
 138                 return -ESPIPE;
 139
 140         /* Null write succeeds. */
 141         if (unlikely(count == 0))
 142                 return 0;
 143
 144         do_wakeup = 0;
 145         ret = 0;
 146         min = count;
 147         if (min > PIPE_BUF)
 148                 min = 1;
 149         down(PIPE_SEM(*inode));
 150         for (;;) {
 151                 int free;
 152                 if (!PIPE_READERS(*inode)) {
 153                         send_sig(SIGPIPE, current, 0);
 154                         if (!ret) ret = -EPIPE;
 155                         break;
 156                 }
 157                 free = PIPE_FREE(*inode);
 158                 if (free >= min) {
 159                         /* transfer data */
 160                         ssize_t chars = PIPE_MAX_WCHUNK(*inode);
 161                         char *pipebuf = PIPE_BASE(*inode) + PIPE_END(*inode);
 162                         /* Always wakeup, even if the copy fails. Otherwise
 163                          * we lock up (O_NONBLOCK-)readers that sleep due to
 164                          * syscall merging.
 165                          */
 166                         do_wakeup = 1;
 167                         if (chars > count)
 168                                 chars = count;
 169                         if (chars > free)
 170                                 chars = free;
 171
 172                         if (copy_from_user(pipebuf, buf, chars)) {
 173                                 if (!ret) ret = -EFAULT;
 174                                 break;
 175                         }
 176
 177                         ret += chars;
 178                         PIPE_LEN(*inode) += chars;
 179                         count -= chars;
 180                         buf += chars;
 181                 }
 182                 if (!count)
 183                         break;
 184                 if (PIPE_FREE(*inode) && ret) {
 185                         /* handle cyclic data buffers */
 186                         min = 1;
 187                         continue;
 188                 }
 189                 if (filp->f_flags & O_NONBLOCK) {
 190                         if (!ret) ret = -EAGAIN;
 191                         break;
 192                 }
 193                 if (signal_pending(current)) {
 194                         if (!ret) ret = -ERESTARTSYS;
 195                         break;
 196                 }
 197                 if (do_wakeup) {
 198                         wake_up_interruptible_sync(PIPE_WAIT(*inode));
 199                         kill_fasync(PIPE_FASYNC_READERS(*inode), SIGIO, POLL_IN);
 200                         do_wakeup = 0;
 201                 }
 202                 PIPE_WAITING_WRITERS(*inode)++;
 203                 pipe_wait(inode);
 204                 PIPE_WAITING_WRITERS(*inode)--;
 205         }
 206         up(PIPE_SEM(*inode));
 207         if (do_wakeup) {
 208                 wake_up_interruptible(PIPE_WAIT(*inode));
 209                 kill_fasync(PIPE_FASYNC_READERS(*inode), SIGIO, POLL_IN);
 210         }
 211         if (ret > 0) {
 212                 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
 213                 mark_inode_dirty(inode);
 214         }
 215         return ret;
 216 }
 217
 218 static ssize_t
 219 bad_pipe_r(struct file *filp, char *buf, size_t count, loff_t *ppos)
 220 {
 221         return -EBADF;
 222 }
 223
 224 static ssize_t
 225 bad_pipe_w(struct file *filp, const char *buf, size_t count, loff_t *ppos)
 226 {
 227         return -EBADF;
 228 }
 229
 230 static int
 231 pipe_ioctl(struct inode *pino, struct file *filp,
 232            unsigned int cmd, unsigned long arg)
 233 {
 234         switch (cmd) {
 235                 case FIONREAD:
 236                         return put_user(PIPE_LEN(*pino), (int *)arg);
 237                 default:
 238                         return -EINVAL;
 239         }
 240 }
 241
 242 /* No kernel lock held - fine */
 243 static unsigned int
 244 pipe_poll(struct file *filp, poll_table *wait)
 245 {
 246         unsigned int mask;
 247         struct inode *inode = filp->f_dentry->d_inode;
 248
 249         poll_wait(filp, PIPE_WAIT(*inode), wait);
 250
 251         /* Reading only -- no need for acquiring the semaphore.  */
 252         mask = POLLIN | POLLRDNORM;
 253         if (PIPE_EMPTY(*inode))
 254                 mask = POLLOUT | POLLWRNORM;
 255         if (!PIPE_WRITERS(*inode) && filp->f_version != PIPE_WCOUNTER(*inode))
 256                 mask |= POLLHUP;
 257         if (!PIPE_READERS(*inode))
 258                 mask |= POLLERR;
 259
 260         return mask;
 261 }
 262
 263 /* FIXME: most Unices do not set POLLERR for fifos */
 264 #define fifo_poll pipe_poll
 265
 266 static int
 267 pipe_release(struct inode *inode, int decr, int decw)
 268 {
 269         down(PIPE_SEM(*inode));
 270         PIPE_READERS(*inode) -= decr;
 271         PIPE_WRITERS(*inode) -= decw;
 272         if (!PIPE_READERS(*inode) && !PIPE_WRITERS(*inode)) {
 273                 struct pipe_inode_info *info = inode->i_pipe;
 274                 inode->i_pipe = NULL;
 275                 free_page((unsigned long) info->base);
 276                 kfree(info);
 277         } else {
 278                 wake_up_interruptible(PIPE_WAIT(*inode));
 279                 kill_fasync(PIPE_FASYNC_READERS(*inode), SIGIO, POLL_IN);
 280                 kill_fasync(PIPE_FASYNC_WRITERS(*inode), SIGIO, POLL_OUT);
 281         }
 282         up(PIPE_SEM(*inode));
 283
 284         return 0;
 285 }
 286
 287 static int
 288 pipe_read_fasync(int fd, struct file *filp, int on)
 289 {
 290         struct inode *inode = filp->f_dentry->d_inode;
 291         int retval;
 292
 293         down(PIPE_SEM(*inode));
 294         retval = fasync_helper(fd, filp, on, PIPE_FASYNC_READERS(*inode));
 295         up(PIPE_SEM(*inode));
 296
 297         if (retval < 0)
 298                 return retval;
 299
 300         return 0;
 301 }
 302
 303
 304 static int
 305 pipe_write_fasync(int fd, struct file *filp, int on)
 306 {
 307         struct inode *inode = filp->f_dentry->d_inode;
 308         int retval;
 309
 310         down(PIPE_SEM(*inode));
 311         retval = fasync_helper(fd, filp, on, PIPE_FASYNC_WRITERS(*inode));
 312         up(PIPE_SEM(*inode));
 313
 314         if (retval < 0)
 315                 return retval;
 316
 317         return 0;
 318 }
 319
 320
 321 static int
 322 pipe_rdwr_fasync(int fd, struct file *filp, int on)
 323 {
 324         struct inode *inode = filp->f_dentry->d_inode;
 325         int retval;
 326
 327         down(PIPE_SEM(*inode));
 328
 329         retval = fasync_helper(fd, filp, on, PIPE_FASYNC_READERS(*inode));
 330
 331         if (retval >= 0)
 332                 retval = fasync_helper(fd, filp, on, PIPE_FASYNC_WRITERS(*inode));
 333
 334         up(PIPE_SEM(*inode));
 335
 336         if (retval < 0)
 337                 return retval;
 338
 339         return 0;
 340 }
 341
 342
 343 static int
 344 pipe_read_release(struct inode *inode, struct file *filp)
 345 {
 346         pipe_read_fasync(-1, filp, 0);
 347         return pipe_release(inode, 1, 0);
 348 }
 349
 350 static int
 351 pipe_write_release(struct inode *inode, struct file *filp)
 352 {
 353         pipe_write_fasync(-1, filp, 0);
 354         return pipe_release(inode, 0, 1);
 355 }
 356
 357 static int
 358 pipe_rdwr_release(struct inode *inode, struct file *filp)
 359 {
 360         int decr, decw;
 361
 362         pipe_rdwr_fasync(-1, filp, 0);
 363         decr = (filp->f_mode & FMODE_READ) != 0;
 364         decw = (filp->f_mode & FMODE_WRITE) != 0;
 365         return pipe_release(inode, decr, decw);
 366 }
 367
 368 static int
 369 pipe_read_open(struct inode *inode, struct file *filp)
 370 {
 371         /* We could have perhaps used atomic_t, but this and friends
 372            below are the only places.  So it doesn't seem worthwhile.  */
 373         down(PIPE_SEM(*inode));
 374         PIPE_READERS(*inode)++;
 375         up(PIPE_SEM(*inode));
 376
 377         return 0;
 378 }
 379
 380 static int
 381 pipe_write_open(struct inode *inode, struct file *filp)
 382 {
 383         down(PIPE_SEM(*inode));
 384         PIPE_WRITERS(*inode)++;
 385         up(PIPE_SEM(*inode));
 386
 387         return 0;
 388 }
 389
 390 static int
 391 pipe_rdwr_open(struct inode *inode, struct file *filp)
 392 {
 393         down(PIPE_SEM(*inode));
 394         if (filp->f_mode & FMODE_READ)
 395                 PIPE_READERS(*inode)++;
 396         if (filp->f_mode & FMODE_WRITE)
 397                 PIPE_WRITERS(*inode)++;
 398         up(PIPE_SEM(*inode));
 399
 400         return 0;
 401 }
 402
 403 /*
 404  * The file_operations structs are not static because they
 405  * are also used in linux/fs/fifo.c to do operations on FIFOs.
 406  */
 407 struct file_operations read_fifo_fops = {
 408         .llseek         = no_llseek,
 409         .read           = pipe_read,
 410         .write          = bad_pipe_w,
 411         .poll           = fifo_poll,
 412         .ioctl          = pipe_ioctl,
 413         .open           = pipe_read_open,
 414         .release        = pipe_read_release,
 415         .fasync         = pipe_read_fasync,
 416 };
 417
 418 struct file_operations write_fifo_fops = {
 419         .llseek         = no_llseek,
 420         .read           = bad_pipe_r,
 421         .write          = pipe_write,
 422         .poll           = fifo_poll,
 423         .ioctl          = pipe_ioctl,
 424         .open           = pipe_write_open,
 425         .release        = pipe_write_release,
 426         .fasync         = pipe_write_fasync,
 427 };
 428
 429 struct file_operations rdwr_fifo_fops = {
 430         .llseek         = no_llseek,
 431         .read           = pipe_read,
 432         .write          = pipe_write,
 433         .poll           = fifo_poll,
 434         .ioctl          = pipe_ioctl,
 435         .open           = pipe_rdwr_open,
 436         .release        = pipe_rdwr_release,
 437         .fasync         = pipe_rdwr_fasync,
 438 };
 439
 440 struct file_operations read_pipe_fops = {
 441         .llseek         = no_llseek,
 442         .read           = pipe_read,
 443         .write          = bad_pipe_w,
 444         .poll           = pipe_poll,
 445         .ioctl          = pipe_ioctl,
 446         .open           = pipe_read_open,
 447         .release        = pipe_read_release,
 448         .fasync         = pipe_read_fasync,
 449 };
 450
 451 struct file_operations write_pipe_fops = {
 452         .llseek         = no_llseek,
 453         .read           = bad_pipe_r,
 454         .write          = pipe_write,
 455         .poll           = pipe_poll,
 456         .ioctl          = pipe_ioctl,
 457         .open           = pipe_write_open,
 458         .release        = pipe_write_release,
 459         .fasync         = pipe_write_fasync,
 460 };
 461
 462 struct file_operations rdwr_pipe_fops = {
 463         .llseek         = no_llseek,
 464         .read           = pipe_read,
 465         .write          = pipe_write,
 466         .poll           = pipe_poll,
 467         .ioctl          = pipe_ioctl,
 468         .open           = pipe_rdwr_open,
 469         .release        = pipe_rdwr_release,
 470         .fasync         = pipe_rdwr_fasync,
 471 };
 472
 473 struct inode* pipe_new(struct inode* inode)
 474 {
 475         unsigned long page;
 476
 477         page = __get_free_page(GFP_USER);
 478         if (!page)
 479                 return NULL;
 480
 481         inode->i_pipe = kmalloc(sizeof(struct pipe_inode_info), GFP_KERNEL);
 482         if (!inode->i_pipe)
 483                 goto fail_page;
 484
 485         init_waitqueue_head(PIPE_WAIT(*inode));
 486         PIPE_BASE(*inode) = (char*) page;
 487         PIPE_START(*inode) = PIPE_LEN(*inode) = 0;
 488         PIPE_READERS(*inode) = PIPE_WRITERS(*inode) = 0;
 489         PIPE_WAITING_WRITERS(*inode) = 0;
 490         PIPE_RCOUNTER(*inode) = PIPE_WCOUNTER(*inode) = 1;
 491         *PIPE_FASYNC_READERS(*inode) = *PIPE_FASYNC_WRITERS(*inode) = NULL;
 492
 493         return inode;
 494 fail_page:
 495         free_page(page);
 496         return NULL;
 497 }
 498
 499 static struct vfsmount *pipe_mnt;
 500 static int pipefs_delete_dentry(struct dentry *dentry)
 501 {
 502         return 1;
 503 }
 504 static struct dentry_operations pipefs_dentry_operations = {
 505         .d_delete       = pipefs_delete_dentry,
 506 };
 507
 508 static struct inode * get_pipe_inode(void)
 509 {
 510         struct inode *inode = new_inode(pipe_mnt->mnt_sb);
 511
 512         if (!inode)
 513                 goto fail_inode;
 514
 515         if(!pipe_new(inode))
 516                 goto fail_iput;
 517         PIPE_READERS(*inode) = PIPE_WRITERS(*inode) = 1;
 518         inode->i_fop = &rdwr_pipe_fops;
 519
 520         /*
 521          * Mark the inode dirty from the very beginning,
 522          * that way it will never be moved to the dirty
 523          * list because "mark_inode_dirty()" will think
 524          * that it already _is_ on the dirty list.
 525          */
 526         inode->i_state = I_DIRTY;
 527         inode->i_mode = S_IFIFO | S_IRUSR | S_IWUSR;
 528         inode->i_uid = current->fsuid;
 529         inode->i_gid = current->fsgid;
 530         inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
 531         inode->i_blksize = PAGE_SIZE;
 532         return inode;
 533
 534 fail_iput:
 535         iput(inode);
 536 fail_inode:
 537         return NULL;
 538 }
 539
 540 int do_pipe(int *fd)
 541 {
 542         struct qstr this;
 543         char name[32];
 544         struct dentry *dentry;
 545         struct inode * inode;
 546         struct file *f1, *f2;
 547         int error;
 548         int i,j;
 549
 550         error = -ENFILE;
 551         f1 = get_empty_filp();
 552         if (!f1)
 553                 goto no_files;
 554
 555         f2 = get_empty_filp();
 556         if (!f2)
 557                 goto close_f1;
 558
 559         inode = get_pipe_inode();
 560         if (!inode)
 561                 goto close_f12;
 562
 563         error = get_unused_fd();
 564         if (error < 0)
 565                 goto close_f12_inode;
 566         i = error;
 567
 568         error = get_unused_fd();
 569         if (error < 0)
 570                 goto close_f12_inode_i;
 571         j = error;
 572
 573         error = -ENOMEM;
 574         sprintf(name, "[%lu]", inode->i_ino);
 575         this.name = name;
 576         this.len = strlen(name);
 577         this.hash = inode->i_ino; /* will go */
 578         dentry = d_alloc(pipe_mnt->mnt_sb->s_root, &this);
 579         if (!dentry)
 580                 goto close_f12_inode_i_j;
 581         dentry->d_op = &pipefs_dentry_operations;
 582         d_add(dentry, inode);
 583         f1->f_vfsmnt = f2->f_vfsmnt = mntget(mntget(pipe_mnt));
 584         f1->f_dentry = f2->f_dentry = dget(dentry);
 585
 586         /* read file */
 587         f1->f_pos = f2->f_pos = 0;
 588         f1->f_flags = O_RDONLY;
 589         f1->f_op = &read_pipe_fops;
 590         f1->f_mode = 1;
 591         f1->f_version = 0;
 592
 593         /* write file */
 594         f2->f_flags = O_WRONLY;
 595         f2->f_op = &write_pipe_fops;
 596         f2->f_mode = 2;
 597         f2->f_version = 0;
 598
 599         fd_install(i, f1);
 600         fd_install(j, f2);
 601         fd[0] = i;
 602         fd[1] = j;
 603         return 0;
 604
 605 close_f12_inode_i_j:
 606         put_unused_fd(j);
 607 close_f12_inode_i:
 608         put_unused_fd(i);
 609 close_f12_inode:
 610         free_page((unsigned long) PIPE_BASE(*inode));
 611         kfree(inode->i_pipe);
 612         inode->i_pipe = NULL;
 613         iput(inode);
 614 close_f12:
 615         put_filp(f2);
 616 close_f1:
 617         put_filp(f1);
 618 no_files:
 619         return error;
 620 }
 621
 622 /*
 623  * pipefs should _never_ be mounted by userland - too much of security hassle,
 624  * no real gain from having the whole whorehouse mounted. So we don't need
 625  * any operations on the root directory. However, we need a non-trivial
 626  * d_name - pipe: will go nicely and kill the special-casing in procfs.
 627  */
 628
 629 static struct super_block *pipefs_get_sb(struct file_system_type *fs_type,
 630         int flags, char *dev_name, void *data)
 631 {
 632         return get_sb_pseudo(fs_type, "pipe:", NULL, PIPEFS_MAGIC);
 633 }
 634
 635 static struct file_system_type pipe_fs_type = {
 636         .name           = "pipefs",
 637         .get_sb         = pipefs_get_sb,
 638         .kill_sb        = kill_anon_super,
 639 };
 640
 641 static int __init init_pipe_fs(void)
 642 {
 643         int err = register_filesystem(&pipe_fs_type);
 644         if (!err) {
 645                 pipe_mnt = kern_mount(&pipe_fs_type);
 646                 err = PTR_ERR(pipe_mnt);
 647                 if (IS_ERR(pipe_mnt))
 648                         unregister_filesystem(&pipe_fs_type);
 649                 else
 650                         err = 0;
 651         }
 652         return err;
 653 }
 654
 655 static void __exit exit_pipe_fs(void)
 656 {
 657         unregister_filesystem(&pipe_fs_type);
 658         mntput(pipe_mnt);
 659 }
 660
 661 module_init(init_pipe_fs)
 662 module_exit(exit_pipe_fs)