Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 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 | |
| 19 | #include <asm/uaccess.h> |
| 20 | #include <asm/ioctls.h> |
| 21 | |
| 22 | /* |
| 23 | * We use a start+len construction, which provides full use of the |
| 24 | * allocated memory. |
| 25 | * -- Florian Coosmann (FGC) |
| 26 | * |
| 27 | * Reads with count = 0 should always return 0. |
| 28 | * -- Julian Bradfield 1999-06-07. |
| 29 | * |
| 30 | * FIFOs and Pipes now generate SIGIO for both readers and writers. |
| 31 | * -- Jeremy Elson <jelson@circlemud.org> 2001-08-16 |
| 32 | * |
| 33 | * pipe_read & write cleanup |
| 34 | * -- Manfred Spraul <manfred@colorfullife.com> 2002-05-09 |
| 35 | */ |
| 36 | |
| 37 | /* Drop the inode semaphore and wait for a pipe event, atomically */ |
| 38 | void pipe_wait(struct inode * inode) |
| 39 | { |
| 40 | DEFINE_WAIT(wait); |
| 41 | |
Ingo Molnar | d79fc0f | 2005-09-10 00:26:12 -0700 | [diff] [blame] | 42 | /* |
| 43 | * Pipes are system-local resources, so sleeping on them |
| 44 | * is considered a noninteractive wait: |
| 45 | */ |
| 46 | prepare_to_wait(PIPE_WAIT(*inode), &wait, TASK_INTERRUPTIBLE|TASK_NONINTERACTIVE); |
Jes Sorensen | 1b1dcc1 | 2006-01-09 15:59:24 -0800 | [diff] [blame] | 47 | mutex_unlock(PIPE_MUTEX(*inode)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 48 | schedule(); |
| 49 | finish_wait(PIPE_WAIT(*inode), &wait); |
Jes Sorensen | 1b1dcc1 | 2006-01-09 15:59:24 -0800 | [diff] [blame] | 50 | mutex_lock(PIPE_MUTEX(*inode)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 51 | } |
| 52 | |
| 53 | static inline int |
| 54 | pipe_iov_copy_from_user(void *to, struct iovec *iov, unsigned long len) |
| 55 | { |
| 56 | unsigned long copy; |
| 57 | |
| 58 | while (len > 0) { |
| 59 | while (!iov->iov_len) |
| 60 | iov++; |
| 61 | copy = min_t(unsigned long, len, iov->iov_len); |
| 62 | |
| 63 | if (copy_from_user(to, iov->iov_base, copy)) |
| 64 | return -EFAULT; |
| 65 | to += copy; |
| 66 | len -= copy; |
| 67 | iov->iov_base += copy; |
| 68 | iov->iov_len -= copy; |
| 69 | } |
| 70 | return 0; |
| 71 | } |
| 72 | |
| 73 | static inline int |
| 74 | pipe_iov_copy_to_user(struct iovec *iov, const void *from, unsigned long len) |
| 75 | { |
| 76 | unsigned long copy; |
| 77 | |
| 78 | while (len > 0) { |
| 79 | while (!iov->iov_len) |
| 80 | iov++; |
| 81 | copy = min_t(unsigned long, len, iov->iov_len); |
| 82 | |
| 83 | if (copy_to_user(iov->iov_base, from, copy)) |
| 84 | return -EFAULT; |
| 85 | from += copy; |
| 86 | len -= copy; |
| 87 | iov->iov_base += copy; |
| 88 | iov->iov_len -= copy; |
| 89 | } |
| 90 | return 0; |
| 91 | } |
| 92 | |
| 93 | static void anon_pipe_buf_release(struct pipe_inode_info *info, struct pipe_buffer *buf) |
| 94 | { |
| 95 | struct page *page = buf->page; |
| 96 | |
| 97 | if (info->tmp_page) { |
| 98 | __free_page(page); |
| 99 | return; |
| 100 | } |
| 101 | info->tmp_page = page; |
| 102 | } |
| 103 | |
| 104 | static void *anon_pipe_buf_map(struct file *file, struct pipe_inode_info *info, struct pipe_buffer *buf) |
| 105 | { |
| 106 | return kmap(buf->page); |
| 107 | } |
| 108 | |
| 109 | static void anon_pipe_buf_unmap(struct pipe_inode_info *info, struct pipe_buffer *buf) |
| 110 | { |
| 111 | kunmap(buf->page); |
| 112 | } |
| 113 | |
| 114 | static struct pipe_buf_operations anon_pipe_buf_ops = { |
| 115 | .can_merge = 1, |
| 116 | .map = anon_pipe_buf_map, |
| 117 | .unmap = anon_pipe_buf_unmap, |
| 118 | .release = anon_pipe_buf_release, |
| 119 | }; |
| 120 | |
| 121 | static ssize_t |
| 122 | pipe_readv(struct file *filp, const struct iovec *_iov, |
| 123 | unsigned long nr_segs, loff_t *ppos) |
| 124 | { |
| 125 | struct inode *inode = filp->f_dentry->d_inode; |
| 126 | struct pipe_inode_info *info; |
| 127 | int do_wakeup; |
| 128 | ssize_t ret; |
| 129 | struct iovec *iov = (struct iovec *)_iov; |
| 130 | size_t total_len; |
| 131 | |
| 132 | total_len = iov_length(iov, nr_segs); |
| 133 | /* Null read succeeds. */ |
| 134 | if (unlikely(total_len == 0)) |
| 135 | return 0; |
| 136 | |
| 137 | do_wakeup = 0; |
| 138 | ret = 0; |
Jes Sorensen | 1b1dcc1 | 2006-01-09 15:59:24 -0800 | [diff] [blame] | 139 | mutex_lock(PIPE_MUTEX(*inode)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 140 | info = inode->i_pipe; |
| 141 | for (;;) { |
| 142 | int bufs = info->nrbufs; |
| 143 | if (bufs) { |
| 144 | int curbuf = info->curbuf; |
| 145 | struct pipe_buffer *buf = info->bufs + curbuf; |
| 146 | struct pipe_buf_operations *ops = buf->ops; |
| 147 | void *addr; |
| 148 | size_t chars = buf->len; |
| 149 | int error; |
| 150 | |
| 151 | if (chars > total_len) |
| 152 | chars = total_len; |
| 153 | |
| 154 | addr = ops->map(filp, info, buf); |
| 155 | error = pipe_iov_copy_to_user(iov, addr + buf->offset, chars); |
| 156 | ops->unmap(info, buf); |
| 157 | if (unlikely(error)) { |
| 158 | if (!ret) ret = -EFAULT; |
| 159 | break; |
| 160 | } |
| 161 | ret += chars; |
| 162 | buf->offset += chars; |
| 163 | buf->len -= chars; |
| 164 | if (!buf->len) { |
| 165 | buf->ops = NULL; |
| 166 | ops->release(info, buf); |
| 167 | curbuf = (curbuf + 1) & (PIPE_BUFFERS-1); |
| 168 | info->curbuf = curbuf; |
| 169 | info->nrbufs = --bufs; |
| 170 | do_wakeup = 1; |
| 171 | } |
| 172 | total_len -= chars; |
| 173 | if (!total_len) |
| 174 | break; /* common path: read succeeded */ |
| 175 | } |
| 176 | if (bufs) /* More to do? */ |
| 177 | continue; |
| 178 | if (!PIPE_WRITERS(*inode)) |
| 179 | break; |
| 180 | if (!PIPE_WAITING_WRITERS(*inode)) { |
| 181 | /* syscall merging: Usually we must not sleep |
| 182 | * if O_NONBLOCK is set, or if we got some data. |
| 183 | * But if a writer sleeps in kernel space, then |
| 184 | * we can wait for that data without violating POSIX. |
| 185 | */ |
| 186 | if (ret) |
| 187 | break; |
| 188 | if (filp->f_flags & O_NONBLOCK) { |
| 189 | ret = -EAGAIN; |
| 190 | break; |
| 191 | } |
| 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_WRITERS(*inode), SIGIO, POLL_OUT); |
| 200 | } |
| 201 | pipe_wait(inode); |
| 202 | } |
Jes Sorensen | 1b1dcc1 | 2006-01-09 15:59:24 -0800 | [diff] [blame] | 203 | mutex_unlock(PIPE_MUTEX(*inode)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 204 | /* Signal writers asynchronously that there is more room. */ |
| 205 | if (do_wakeup) { |
| 206 | wake_up_interruptible(PIPE_WAIT(*inode)); |
| 207 | kill_fasync(PIPE_FASYNC_WRITERS(*inode), SIGIO, POLL_OUT); |
| 208 | } |
| 209 | if (ret > 0) |
| 210 | file_accessed(filp); |
| 211 | return ret; |
| 212 | } |
| 213 | |
| 214 | static ssize_t |
| 215 | pipe_read(struct file *filp, char __user *buf, size_t count, loff_t *ppos) |
| 216 | { |
| 217 | struct iovec iov = { .iov_base = buf, .iov_len = count }; |
| 218 | return pipe_readv(filp, &iov, 1, ppos); |
| 219 | } |
| 220 | |
| 221 | static ssize_t |
| 222 | pipe_writev(struct file *filp, const struct iovec *_iov, |
| 223 | unsigned long nr_segs, loff_t *ppos) |
| 224 | { |
| 225 | struct inode *inode = filp->f_dentry->d_inode; |
| 226 | struct pipe_inode_info *info; |
| 227 | ssize_t ret; |
| 228 | int do_wakeup; |
| 229 | struct iovec *iov = (struct iovec *)_iov; |
| 230 | size_t total_len; |
| 231 | ssize_t chars; |
| 232 | |
| 233 | total_len = iov_length(iov, nr_segs); |
| 234 | /* Null write succeeds. */ |
| 235 | if (unlikely(total_len == 0)) |
| 236 | return 0; |
| 237 | |
| 238 | do_wakeup = 0; |
| 239 | ret = 0; |
Jes Sorensen | 1b1dcc1 | 2006-01-09 15:59:24 -0800 | [diff] [blame] | 240 | mutex_lock(PIPE_MUTEX(*inode)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 241 | info = inode->i_pipe; |
| 242 | |
| 243 | if (!PIPE_READERS(*inode)) { |
| 244 | send_sig(SIGPIPE, current, 0); |
| 245 | ret = -EPIPE; |
| 246 | goto out; |
| 247 | } |
| 248 | |
| 249 | /* We try to merge small writes */ |
| 250 | chars = total_len & (PAGE_SIZE-1); /* size of the last buffer */ |
| 251 | if (info->nrbufs && chars != 0) { |
| 252 | int lastbuf = (info->curbuf + info->nrbufs - 1) & (PIPE_BUFFERS-1); |
| 253 | struct pipe_buffer *buf = info->bufs + lastbuf; |
| 254 | struct pipe_buf_operations *ops = buf->ops; |
| 255 | int offset = buf->offset + buf->len; |
| 256 | if (ops->can_merge && offset + chars <= PAGE_SIZE) { |
| 257 | void *addr = ops->map(filp, info, buf); |
| 258 | int error = pipe_iov_copy_from_user(offset + addr, iov, chars); |
| 259 | ops->unmap(info, buf); |
| 260 | ret = error; |
| 261 | do_wakeup = 1; |
| 262 | if (error) |
| 263 | goto out; |
| 264 | buf->len += chars; |
| 265 | total_len -= chars; |
| 266 | ret = chars; |
| 267 | if (!total_len) |
| 268 | goto out; |
| 269 | } |
| 270 | } |
| 271 | |
| 272 | for (;;) { |
| 273 | int bufs; |
| 274 | if (!PIPE_READERS(*inode)) { |
| 275 | send_sig(SIGPIPE, current, 0); |
| 276 | if (!ret) ret = -EPIPE; |
| 277 | break; |
| 278 | } |
| 279 | bufs = info->nrbufs; |
| 280 | if (bufs < PIPE_BUFFERS) { |
| 281 | int newbuf = (info->curbuf + bufs) & (PIPE_BUFFERS-1); |
| 282 | struct pipe_buffer *buf = info->bufs + newbuf; |
| 283 | struct page *page = info->tmp_page; |
| 284 | int error; |
| 285 | |
| 286 | if (!page) { |
| 287 | page = alloc_page(GFP_HIGHUSER); |
| 288 | if (unlikely(!page)) { |
| 289 | ret = ret ? : -ENOMEM; |
| 290 | break; |
| 291 | } |
| 292 | info->tmp_page = page; |
| 293 | } |
| 294 | /* Always wakeup, even if the copy fails. Otherwise |
| 295 | * we lock up (O_NONBLOCK-)readers that sleep due to |
| 296 | * syscall merging. |
| 297 | * FIXME! Is this really true? |
| 298 | */ |
| 299 | do_wakeup = 1; |
| 300 | chars = PAGE_SIZE; |
| 301 | if (chars > total_len) |
| 302 | chars = total_len; |
| 303 | |
| 304 | error = pipe_iov_copy_from_user(kmap(page), iov, chars); |
| 305 | kunmap(page); |
| 306 | if (unlikely(error)) { |
| 307 | if (!ret) ret = -EFAULT; |
| 308 | break; |
| 309 | } |
| 310 | ret += chars; |
| 311 | |
| 312 | /* Insert it into the buffer array */ |
| 313 | buf->page = page; |
| 314 | buf->ops = &anon_pipe_buf_ops; |
| 315 | buf->offset = 0; |
| 316 | buf->len = chars; |
| 317 | info->nrbufs = ++bufs; |
| 318 | info->tmp_page = NULL; |
| 319 | |
| 320 | total_len -= chars; |
| 321 | if (!total_len) |
| 322 | break; |
| 323 | } |
| 324 | if (bufs < PIPE_BUFFERS) |
| 325 | continue; |
| 326 | if (filp->f_flags & O_NONBLOCK) { |
| 327 | if (!ret) ret = -EAGAIN; |
| 328 | break; |
| 329 | } |
| 330 | if (signal_pending(current)) { |
| 331 | if (!ret) ret = -ERESTARTSYS; |
| 332 | break; |
| 333 | } |
| 334 | if (do_wakeup) { |
| 335 | wake_up_interruptible_sync(PIPE_WAIT(*inode)); |
| 336 | kill_fasync(PIPE_FASYNC_READERS(*inode), SIGIO, POLL_IN); |
| 337 | do_wakeup = 0; |
| 338 | } |
| 339 | PIPE_WAITING_WRITERS(*inode)++; |
| 340 | pipe_wait(inode); |
| 341 | PIPE_WAITING_WRITERS(*inode)--; |
| 342 | } |
| 343 | out: |
Jes Sorensen | 1b1dcc1 | 2006-01-09 15:59:24 -0800 | [diff] [blame] | 344 | mutex_unlock(PIPE_MUTEX(*inode)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 345 | if (do_wakeup) { |
| 346 | wake_up_interruptible(PIPE_WAIT(*inode)); |
| 347 | kill_fasync(PIPE_FASYNC_READERS(*inode), SIGIO, POLL_IN); |
| 348 | } |
| 349 | if (ret > 0) |
Christoph Hellwig | 870f481 | 2006-01-09 20:52:01 -0800 | [diff] [blame] | 350 | file_update_time(filp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 351 | return ret; |
| 352 | } |
| 353 | |
| 354 | static ssize_t |
| 355 | pipe_write(struct file *filp, const char __user *buf, |
| 356 | size_t count, loff_t *ppos) |
| 357 | { |
| 358 | struct iovec iov = { .iov_base = (void __user *)buf, .iov_len = count }; |
| 359 | return pipe_writev(filp, &iov, 1, ppos); |
| 360 | } |
| 361 | |
| 362 | static ssize_t |
| 363 | bad_pipe_r(struct file *filp, char __user *buf, size_t count, loff_t *ppos) |
| 364 | { |
| 365 | return -EBADF; |
| 366 | } |
| 367 | |
| 368 | static ssize_t |
| 369 | bad_pipe_w(struct file *filp, const char __user *buf, size_t count, loff_t *ppos) |
| 370 | { |
| 371 | return -EBADF; |
| 372 | } |
| 373 | |
| 374 | static int |
| 375 | pipe_ioctl(struct inode *pino, struct file *filp, |
| 376 | unsigned int cmd, unsigned long arg) |
| 377 | { |
| 378 | struct inode *inode = filp->f_dentry->d_inode; |
| 379 | struct pipe_inode_info *info; |
| 380 | int count, buf, nrbufs; |
| 381 | |
| 382 | switch (cmd) { |
| 383 | case FIONREAD: |
Jes Sorensen | 1b1dcc1 | 2006-01-09 15:59:24 -0800 | [diff] [blame] | 384 | mutex_lock(PIPE_MUTEX(*inode)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 385 | info = inode->i_pipe; |
| 386 | count = 0; |
| 387 | buf = info->curbuf; |
| 388 | nrbufs = info->nrbufs; |
| 389 | while (--nrbufs >= 0) { |
| 390 | count += info->bufs[buf].len; |
| 391 | buf = (buf+1) & (PIPE_BUFFERS-1); |
| 392 | } |
Jes Sorensen | 1b1dcc1 | 2006-01-09 15:59:24 -0800 | [diff] [blame] | 393 | mutex_unlock(PIPE_MUTEX(*inode)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 394 | return put_user(count, (int __user *)arg); |
| 395 | default: |
| 396 | return -EINVAL; |
| 397 | } |
| 398 | } |
| 399 | |
| 400 | /* No kernel lock held - fine */ |
| 401 | static unsigned int |
| 402 | pipe_poll(struct file *filp, poll_table *wait) |
| 403 | { |
| 404 | unsigned int mask; |
| 405 | struct inode *inode = filp->f_dentry->d_inode; |
| 406 | struct pipe_inode_info *info = inode->i_pipe; |
| 407 | int nrbufs; |
| 408 | |
| 409 | poll_wait(filp, PIPE_WAIT(*inode), wait); |
| 410 | |
| 411 | /* Reading only -- no need for acquiring the semaphore. */ |
| 412 | nrbufs = info->nrbufs; |
| 413 | mask = 0; |
| 414 | if (filp->f_mode & FMODE_READ) { |
| 415 | mask = (nrbufs > 0) ? POLLIN | POLLRDNORM : 0; |
| 416 | if (!PIPE_WRITERS(*inode) && filp->f_version != PIPE_WCOUNTER(*inode)) |
| 417 | mask |= POLLHUP; |
| 418 | } |
| 419 | |
| 420 | if (filp->f_mode & FMODE_WRITE) { |
| 421 | mask |= (nrbufs < PIPE_BUFFERS) ? POLLOUT | POLLWRNORM : 0; |
Pekka Enberg | 5e5d7a2 | 2005-09-06 15:17:48 -0700 | [diff] [blame] | 422 | /* |
| 423 | * Most Unices do not set POLLERR for FIFOs but on Linux they |
| 424 | * behave exactly like pipes for poll(). |
| 425 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 426 | if (!PIPE_READERS(*inode)) |
| 427 | mask |= POLLERR; |
| 428 | } |
| 429 | |
| 430 | return mask; |
| 431 | } |
| 432 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 433 | static int |
| 434 | pipe_release(struct inode *inode, int decr, int decw) |
| 435 | { |
Jes Sorensen | 1b1dcc1 | 2006-01-09 15:59:24 -0800 | [diff] [blame] | 436 | mutex_lock(PIPE_MUTEX(*inode)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 437 | PIPE_READERS(*inode) -= decr; |
| 438 | PIPE_WRITERS(*inode) -= decw; |
| 439 | if (!PIPE_READERS(*inode) && !PIPE_WRITERS(*inode)) { |
| 440 | free_pipe_info(inode); |
| 441 | } else { |
| 442 | wake_up_interruptible(PIPE_WAIT(*inode)); |
| 443 | kill_fasync(PIPE_FASYNC_READERS(*inode), SIGIO, POLL_IN); |
| 444 | kill_fasync(PIPE_FASYNC_WRITERS(*inode), SIGIO, POLL_OUT); |
| 445 | } |
Jes Sorensen | 1b1dcc1 | 2006-01-09 15:59:24 -0800 | [diff] [blame] | 446 | mutex_unlock(PIPE_MUTEX(*inode)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 447 | |
| 448 | return 0; |
| 449 | } |
| 450 | |
| 451 | static int |
| 452 | pipe_read_fasync(int fd, struct file *filp, int on) |
| 453 | { |
| 454 | struct inode *inode = filp->f_dentry->d_inode; |
| 455 | int retval; |
| 456 | |
Jes Sorensen | 1b1dcc1 | 2006-01-09 15:59:24 -0800 | [diff] [blame] | 457 | mutex_lock(PIPE_MUTEX(*inode)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 458 | retval = fasync_helper(fd, filp, on, PIPE_FASYNC_READERS(*inode)); |
Jes Sorensen | 1b1dcc1 | 2006-01-09 15:59:24 -0800 | [diff] [blame] | 459 | mutex_unlock(PIPE_MUTEX(*inode)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 460 | |
| 461 | if (retval < 0) |
| 462 | return retval; |
| 463 | |
| 464 | return 0; |
| 465 | } |
| 466 | |
| 467 | |
| 468 | static int |
| 469 | pipe_write_fasync(int fd, struct file *filp, int on) |
| 470 | { |
| 471 | struct inode *inode = filp->f_dentry->d_inode; |
| 472 | int retval; |
| 473 | |
Jes Sorensen | 1b1dcc1 | 2006-01-09 15:59:24 -0800 | [diff] [blame] | 474 | mutex_lock(PIPE_MUTEX(*inode)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 475 | retval = fasync_helper(fd, filp, on, PIPE_FASYNC_WRITERS(*inode)); |
Jes Sorensen | 1b1dcc1 | 2006-01-09 15:59:24 -0800 | [diff] [blame] | 476 | mutex_unlock(PIPE_MUTEX(*inode)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 477 | |
| 478 | if (retval < 0) |
| 479 | return retval; |
| 480 | |
| 481 | return 0; |
| 482 | } |
| 483 | |
| 484 | |
| 485 | static int |
| 486 | pipe_rdwr_fasync(int fd, struct file *filp, int on) |
| 487 | { |
| 488 | struct inode *inode = filp->f_dentry->d_inode; |
| 489 | int retval; |
| 490 | |
Jes Sorensen | 1b1dcc1 | 2006-01-09 15:59:24 -0800 | [diff] [blame] | 491 | mutex_lock(PIPE_MUTEX(*inode)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 492 | |
| 493 | retval = fasync_helper(fd, filp, on, PIPE_FASYNC_READERS(*inode)); |
| 494 | |
| 495 | if (retval >= 0) |
| 496 | retval = fasync_helper(fd, filp, on, PIPE_FASYNC_WRITERS(*inode)); |
| 497 | |
Jes Sorensen | 1b1dcc1 | 2006-01-09 15:59:24 -0800 | [diff] [blame] | 498 | mutex_unlock(PIPE_MUTEX(*inode)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 499 | |
| 500 | if (retval < 0) |
| 501 | return retval; |
| 502 | |
| 503 | return 0; |
| 504 | } |
| 505 | |
| 506 | |
| 507 | static int |
| 508 | pipe_read_release(struct inode *inode, struct file *filp) |
| 509 | { |
| 510 | pipe_read_fasync(-1, filp, 0); |
| 511 | return pipe_release(inode, 1, 0); |
| 512 | } |
| 513 | |
| 514 | static int |
| 515 | pipe_write_release(struct inode *inode, struct file *filp) |
| 516 | { |
| 517 | pipe_write_fasync(-1, filp, 0); |
| 518 | return pipe_release(inode, 0, 1); |
| 519 | } |
| 520 | |
| 521 | static int |
| 522 | pipe_rdwr_release(struct inode *inode, struct file *filp) |
| 523 | { |
| 524 | int decr, decw; |
| 525 | |
| 526 | pipe_rdwr_fasync(-1, filp, 0); |
| 527 | decr = (filp->f_mode & FMODE_READ) != 0; |
| 528 | decw = (filp->f_mode & FMODE_WRITE) != 0; |
| 529 | return pipe_release(inode, decr, decw); |
| 530 | } |
| 531 | |
| 532 | static int |
| 533 | pipe_read_open(struct inode *inode, struct file *filp) |
| 534 | { |
| 535 | /* We could have perhaps used atomic_t, but this and friends |
| 536 | below are the only places. So it doesn't seem worthwhile. */ |
Jes Sorensen | 1b1dcc1 | 2006-01-09 15:59:24 -0800 | [diff] [blame] | 537 | mutex_lock(PIPE_MUTEX(*inode)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 538 | PIPE_READERS(*inode)++; |
Jes Sorensen | 1b1dcc1 | 2006-01-09 15:59:24 -0800 | [diff] [blame] | 539 | mutex_unlock(PIPE_MUTEX(*inode)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 540 | |
| 541 | return 0; |
| 542 | } |
| 543 | |
| 544 | static int |
| 545 | pipe_write_open(struct inode *inode, struct file *filp) |
| 546 | { |
Jes Sorensen | 1b1dcc1 | 2006-01-09 15:59:24 -0800 | [diff] [blame] | 547 | mutex_lock(PIPE_MUTEX(*inode)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 548 | PIPE_WRITERS(*inode)++; |
Jes Sorensen | 1b1dcc1 | 2006-01-09 15:59:24 -0800 | [diff] [blame] | 549 | mutex_unlock(PIPE_MUTEX(*inode)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 550 | |
| 551 | return 0; |
| 552 | } |
| 553 | |
| 554 | static int |
| 555 | pipe_rdwr_open(struct inode *inode, struct file *filp) |
| 556 | { |
Jes Sorensen | 1b1dcc1 | 2006-01-09 15:59:24 -0800 | [diff] [blame] | 557 | mutex_lock(PIPE_MUTEX(*inode)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 558 | if (filp->f_mode & FMODE_READ) |
| 559 | PIPE_READERS(*inode)++; |
| 560 | if (filp->f_mode & FMODE_WRITE) |
| 561 | PIPE_WRITERS(*inode)++; |
Jes Sorensen | 1b1dcc1 | 2006-01-09 15:59:24 -0800 | [diff] [blame] | 562 | mutex_unlock(PIPE_MUTEX(*inode)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 563 | |
| 564 | return 0; |
| 565 | } |
| 566 | |
| 567 | /* |
| 568 | * The file_operations structs are not static because they |
| 569 | * are also used in linux/fs/fifo.c to do operations on FIFOs. |
| 570 | */ |
| 571 | struct file_operations read_fifo_fops = { |
| 572 | .llseek = no_llseek, |
| 573 | .read = pipe_read, |
| 574 | .readv = pipe_readv, |
| 575 | .write = bad_pipe_w, |
Pekka Enberg | 5e5d7a2 | 2005-09-06 15:17:48 -0700 | [diff] [blame] | 576 | .poll = pipe_poll, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 577 | .ioctl = pipe_ioctl, |
| 578 | .open = pipe_read_open, |
| 579 | .release = pipe_read_release, |
| 580 | .fasync = pipe_read_fasync, |
| 581 | }; |
| 582 | |
| 583 | struct file_operations write_fifo_fops = { |
| 584 | .llseek = no_llseek, |
| 585 | .read = bad_pipe_r, |
| 586 | .write = pipe_write, |
| 587 | .writev = pipe_writev, |
Pekka Enberg | 5e5d7a2 | 2005-09-06 15:17:48 -0700 | [diff] [blame] | 588 | .poll = pipe_poll, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 589 | .ioctl = pipe_ioctl, |
| 590 | .open = pipe_write_open, |
| 591 | .release = pipe_write_release, |
| 592 | .fasync = pipe_write_fasync, |
| 593 | }; |
| 594 | |
| 595 | struct file_operations rdwr_fifo_fops = { |
| 596 | .llseek = no_llseek, |
| 597 | .read = pipe_read, |
| 598 | .readv = pipe_readv, |
| 599 | .write = pipe_write, |
| 600 | .writev = pipe_writev, |
Pekka Enberg | 5e5d7a2 | 2005-09-06 15:17:48 -0700 | [diff] [blame] | 601 | .poll = pipe_poll, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 602 | .ioctl = pipe_ioctl, |
| 603 | .open = pipe_rdwr_open, |
| 604 | .release = pipe_rdwr_release, |
| 605 | .fasync = pipe_rdwr_fasync, |
| 606 | }; |
| 607 | |
| 608 | struct file_operations read_pipe_fops = { |
| 609 | .llseek = no_llseek, |
| 610 | .read = pipe_read, |
| 611 | .readv = pipe_readv, |
| 612 | .write = bad_pipe_w, |
| 613 | .poll = pipe_poll, |
| 614 | .ioctl = pipe_ioctl, |
| 615 | .open = pipe_read_open, |
| 616 | .release = pipe_read_release, |
| 617 | .fasync = pipe_read_fasync, |
| 618 | }; |
| 619 | |
| 620 | struct file_operations write_pipe_fops = { |
| 621 | .llseek = no_llseek, |
| 622 | .read = bad_pipe_r, |
| 623 | .write = pipe_write, |
| 624 | .writev = pipe_writev, |
| 625 | .poll = pipe_poll, |
| 626 | .ioctl = pipe_ioctl, |
| 627 | .open = pipe_write_open, |
| 628 | .release = pipe_write_release, |
| 629 | .fasync = pipe_write_fasync, |
| 630 | }; |
| 631 | |
| 632 | struct file_operations rdwr_pipe_fops = { |
| 633 | .llseek = no_llseek, |
| 634 | .read = pipe_read, |
| 635 | .readv = pipe_readv, |
| 636 | .write = pipe_write, |
| 637 | .writev = pipe_writev, |
| 638 | .poll = pipe_poll, |
| 639 | .ioctl = pipe_ioctl, |
| 640 | .open = pipe_rdwr_open, |
| 641 | .release = pipe_rdwr_release, |
| 642 | .fasync = pipe_rdwr_fasync, |
| 643 | }; |
| 644 | |
| 645 | void free_pipe_info(struct inode *inode) |
| 646 | { |
| 647 | int i; |
| 648 | struct pipe_inode_info *info = inode->i_pipe; |
| 649 | |
| 650 | inode->i_pipe = NULL; |
| 651 | for (i = 0; i < PIPE_BUFFERS; i++) { |
| 652 | struct pipe_buffer *buf = info->bufs + i; |
| 653 | if (buf->ops) |
| 654 | buf->ops->release(info, buf); |
| 655 | } |
| 656 | if (info->tmp_page) |
| 657 | __free_page(info->tmp_page); |
| 658 | kfree(info); |
| 659 | } |
| 660 | |
| 661 | struct inode* pipe_new(struct inode* inode) |
| 662 | { |
| 663 | struct pipe_inode_info *info; |
| 664 | |
| 665 | info = kmalloc(sizeof(struct pipe_inode_info), GFP_KERNEL); |
| 666 | if (!info) |
| 667 | goto fail_page; |
| 668 | memset(info, 0, sizeof(*info)); |
| 669 | inode->i_pipe = info; |
| 670 | |
| 671 | init_waitqueue_head(PIPE_WAIT(*inode)); |
| 672 | PIPE_RCOUNTER(*inode) = PIPE_WCOUNTER(*inode) = 1; |
| 673 | |
| 674 | return inode; |
| 675 | fail_page: |
| 676 | return NULL; |
| 677 | } |
| 678 | |
| 679 | static struct vfsmount *pipe_mnt; |
| 680 | static int pipefs_delete_dentry(struct dentry *dentry) |
| 681 | { |
| 682 | return 1; |
| 683 | } |
| 684 | static struct dentry_operations pipefs_dentry_operations = { |
| 685 | .d_delete = pipefs_delete_dentry, |
| 686 | }; |
| 687 | |
| 688 | static struct inode * get_pipe_inode(void) |
| 689 | { |
| 690 | struct inode *inode = new_inode(pipe_mnt->mnt_sb); |
| 691 | |
| 692 | if (!inode) |
| 693 | goto fail_inode; |
| 694 | |
| 695 | if(!pipe_new(inode)) |
| 696 | goto fail_iput; |
| 697 | PIPE_READERS(*inode) = PIPE_WRITERS(*inode) = 1; |
| 698 | inode->i_fop = &rdwr_pipe_fops; |
| 699 | |
| 700 | /* |
| 701 | * Mark the inode dirty from the very beginning, |
| 702 | * that way it will never be moved to the dirty |
| 703 | * list because "mark_inode_dirty()" will think |
| 704 | * that it already _is_ on the dirty list. |
| 705 | */ |
| 706 | inode->i_state = I_DIRTY; |
| 707 | inode->i_mode = S_IFIFO | S_IRUSR | S_IWUSR; |
| 708 | inode->i_uid = current->fsuid; |
| 709 | inode->i_gid = current->fsgid; |
| 710 | inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME; |
| 711 | inode->i_blksize = PAGE_SIZE; |
| 712 | return inode; |
| 713 | |
| 714 | fail_iput: |
| 715 | iput(inode); |
| 716 | fail_inode: |
| 717 | return NULL; |
| 718 | } |
| 719 | |
| 720 | int do_pipe(int *fd) |
| 721 | { |
| 722 | struct qstr this; |
| 723 | char name[32]; |
| 724 | struct dentry *dentry; |
| 725 | struct inode * inode; |
| 726 | struct file *f1, *f2; |
| 727 | int error; |
| 728 | int i,j; |
| 729 | |
| 730 | error = -ENFILE; |
| 731 | f1 = get_empty_filp(); |
| 732 | if (!f1) |
| 733 | goto no_files; |
| 734 | |
| 735 | f2 = get_empty_filp(); |
| 736 | if (!f2) |
| 737 | goto close_f1; |
| 738 | |
| 739 | inode = get_pipe_inode(); |
| 740 | if (!inode) |
| 741 | goto close_f12; |
| 742 | |
| 743 | error = get_unused_fd(); |
| 744 | if (error < 0) |
| 745 | goto close_f12_inode; |
| 746 | i = error; |
| 747 | |
| 748 | error = get_unused_fd(); |
| 749 | if (error < 0) |
| 750 | goto close_f12_inode_i; |
| 751 | j = error; |
| 752 | |
| 753 | error = -ENOMEM; |
| 754 | sprintf(name, "[%lu]", inode->i_ino); |
| 755 | this.name = name; |
| 756 | this.len = strlen(name); |
| 757 | this.hash = inode->i_ino; /* will go */ |
| 758 | dentry = d_alloc(pipe_mnt->mnt_sb->s_root, &this); |
| 759 | if (!dentry) |
| 760 | goto close_f12_inode_i_j; |
| 761 | dentry->d_op = &pipefs_dentry_operations; |
| 762 | d_add(dentry, inode); |
| 763 | f1->f_vfsmnt = f2->f_vfsmnt = mntget(mntget(pipe_mnt)); |
| 764 | f1->f_dentry = f2->f_dentry = dget(dentry); |
| 765 | f1->f_mapping = f2->f_mapping = inode->i_mapping; |
| 766 | |
| 767 | /* read file */ |
| 768 | f1->f_pos = f2->f_pos = 0; |
| 769 | f1->f_flags = O_RDONLY; |
| 770 | f1->f_op = &read_pipe_fops; |
| 771 | f1->f_mode = FMODE_READ; |
| 772 | f1->f_version = 0; |
| 773 | |
| 774 | /* write file */ |
| 775 | f2->f_flags = O_WRONLY; |
| 776 | f2->f_op = &write_pipe_fops; |
| 777 | f2->f_mode = FMODE_WRITE; |
| 778 | f2->f_version = 0; |
| 779 | |
| 780 | fd_install(i, f1); |
| 781 | fd_install(j, f2); |
| 782 | fd[0] = i; |
| 783 | fd[1] = j; |
| 784 | return 0; |
| 785 | |
| 786 | close_f12_inode_i_j: |
| 787 | put_unused_fd(j); |
| 788 | close_f12_inode_i: |
| 789 | put_unused_fd(i); |
| 790 | close_f12_inode: |
| 791 | free_pipe_info(inode); |
| 792 | iput(inode); |
| 793 | close_f12: |
| 794 | put_filp(f2); |
| 795 | close_f1: |
| 796 | put_filp(f1); |
| 797 | no_files: |
| 798 | return error; |
| 799 | } |
| 800 | |
| 801 | /* |
| 802 | * pipefs should _never_ be mounted by userland - too much of security hassle, |
| 803 | * no real gain from having the whole whorehouse mounted. So we don't need |
| 804 | * any operations on the root directory. However, we need a non-trivial |
| 805 | * d_name - pipe: will go nicely and kill the special-casing in procfs. |
| 806 | */ |
| 807 | |
| 808 | static struct super_block *pipefs_get_sb(struct file_system_type *fs_type, |
| 809 | int flags, const char *dev_name, void *data) |
| 810 | { |
| 811 | return get_sb_pseudo(fs_type, "pipe:", NULL, PIPEFS_MAGIC); |
| 812 | } |
| 813 | |
| 814 | static struct file_system_type pipe_fs_type = { |
| 815 | .name = "pipefs", |
| 816 | .get_sb = pipefs_get_sb, |
| 817 | .kill_sb = kill_anon_super, |
| 818 | }; |
| 819 | |
| 820 | static int __init init_pipe_fs(void) |
| 821 | { |
| 822 | int err = register_filesystem(&pipe_fs_type); |
| 823 | if (!err) { |
| 824 | pipe_mnt = kern_mount(&pipe_fs_type); |
| 825 | if (IS_ERR(pipe_mnt)) { |
| 826 | err = PTR_ERR(pipe_mnt); |
| 827 | unregister_filesystem(&pipe_fs_type); |
| 828 | } |
| 829 | } |
| 830 | return err; |
| 831 | } |
| 832 | |
| 833 | static void __exit exit_pipe_fs(void) |
| 834 | { |
| 835 | unregister_filesystem(&pipe_fs_type); |
| 836 | mntput(pipe_mnt); |
| 837 | } |
| 838 | |
| 839 | fs_initcall(init_pipe_fs); |
| 840 | module_exit(exit_pipe_fs); |