SAFE public projects git trees. - safe/jmp/linux-2.6/blob - kernel/relay.c

   1 /*
   2  * Public API and common code for kernel->userspace relay file support.
   3  *
   4  * See Documentation/filesystems/relayfs.txt for an overview of relayfs.
   5  *
   6  * Copyright (C) 2002-2005 - Tom Zanussi (zanussi@us.ibm.com), IBM Corp
   7  * Copyright (C) 1999-2005 - Karim Yaghmour (karim@opersys.com)
   8  *
   9  * Moved to kernel/relay.c by Paul Mundt, 2006.
  10  * November 2006 - CPU hotplug support by Mathieu Desnoyers
  11  *      (mathieu.desnoyers@polymtl.ca)
  12  *
  13  * This file is released under the GPL.
  14  */
  15 #include <linux/errno.h>
  16 #include <linux/stddef.h>
  17 #include <linux/slab.h>
  18 #include <linux/module.h>
  19 #include <linux/string.h>
  20 #include <linux/relay.h>
  21 #include <linux/vmalloc.h>
  22 #include <linux/mm.h>
  23 #include <linux/cpu.h>
  24
  25 /* list of open channels, for cpu hotplug */
  26 static DEFINE_MUTEX(relay_channels_mutex);
  27 static LIST_HEAD(relay_channels);
  28
  29 /*
  30  * close() vm_op implementation for relay file mapping.
  31  */
  32 static void relay_file_mmap_close(struct vm_area_struct *vma)
  33 {
  34         struct rchan_buf *buf = vma->vm_private_data;
  35         buf->chan->cb->buf_unmapped(buf, vma->vm_file);
  36 }
  37
  38 /*
  39  * nopage() vm_op implementation for relay file mapping.
  40  */
  41 static struct page *relay_buf_nopage(struct vm_area_struct *vma,
  42                                      unsigned long address,
  43                                      int *type)
  44 {
  45         struct page *page;
  46         struct rchan_buf *buf = vma->vm_private_data;
  47         unsigned long offset = address - vma->vm_start;
  48
  49         if (address > vma->vm_end)
  50                 return NOPAGE_SIGBUS; /* Disallow mremap */
  51         if (!buf)
  52                 return NOPAGE_OOM;
  53
  54         page = vmalloc_to_page(buf->start + offset);
  55         if (!page)
  56                 return NOPAGE_OOM;
  57         get_page(page);
  58
  59         if (type)
  60                 *type = VM_FAULT_MINOR;
  61
  62         return page;
  63 }
  64
  65 /*
  66  * vm_ops for relay file mappings.
  67  */
  68 static struct vm_operations_struct relay_file_mmap_ops = {
  69         .nopage = relay_buf_nopage,
  70         .close = relay_file_mmap_close,
  71 };
  72
  73 /**
  74  *      relay_mmap_buf: - mmap channel buffer to process address space
  75  *      @buf: relay channel buffer
  76  *      @vma: vm_area_struct describing memory to be mapped
  77  *
  78  *      Returns 0 if ok, negative on error
  79  *
  80  *      Caller should already have grabbed mmap_sem.
  81  */
  82 int relay_mmap_buf(struct rchan_buf *buf, struct vm_area_struct *vma)
  83 {
  84         unsigned long length = vma->vm_end - vma->vm_start;
  85         struct file *filp = vma->vm_file;
  86
  87         if (!buf)
  88                 return -EBADF;
  89
  90         if (length != (unsigned long)buf->chan->alloc_size)
  91                 return -EINVAL;
  92
  93         vma->vm_ops = &relay_file_mmap_ops;
  94         vma->vm_private_data = buf;
  95         buf->chan->cb->buf_mapped(buf, filp);
  96
  97         return 0;
  98 }
  99
 100 /**
 101  *      relay_alloc_buf - allocate a channel buffer
 102  *      @buf: the buffer struct
 103  *      @size: total size of the buffer
 104  *
 105  *      Returns a pointer to the resulting buffer, %NULL if unsuccessful. The
 106  *      passed in size will get page aligned, if it isn't already.
 107  */
 108 static void *relay_alloc_buf(struct rchan_buf *buf, size_t *size)
 109 {
 110         void *mem;
 111         unsigned int i, j, n_pages;
 112
 113         *size = PAGE_ALIGN(*size);
 114         n_pages = *size >> PAGE_SHIFT;
 115
 116         buf->page_array = kcalloc(n_pages, sizeof(struct page *), GFP_KERNEL);
 117         if (!buf->page_array)
 118                 return NULL;
 119
 120         for (i = 0; i < n_pages; i++) {
 121                 buf->page_array[i] = alloc_page(GFP_KERNEL);
 122                 if (unlikely(!buf->page_array[i]))
 123                         goto depopulate;
 124         }
 125         mem = vmap(buf->page_array, n_pages, VM_MAP, PAGE_KERNEL);
 126         if (!mem)
 127                 goto depopulate;
 128
 129         memset(mem, 0, *size);
 130         buf->page_count = n_pages;
 131         return mem;
 132
 133 depopulate:
 134         for (j = 0; j < i; j++)
 135                 __free_page(buf->page_array[j]);
 136         kfree(buf->page_array);
 137         return NULL;
 138 }
 139
 140 /**
 141  *      relay_create_buf - allocate and initialize a channel buffer
 142  *      @chan: the relay channel
 143  *
 144  *      Returns channel buffer if successful, %NULL otherwise.
 145  */
 146 struct rchan_buf *relay_create_buf(struct rchan *chan)
 147 {
 148         struct rchan_buf *buf = kzalloc(sizeof(struct rchan_buf), GFP_KERNEL);
 149         if (!buf)
 150                 return NULL;
 151
 152         buf->padding = kmalloc(chan->n_subbufs * sizeof(size_t *), GFP_KERNEL);
 153         if (!buf->padding)
 154                 goto free_buf;
 155
 156         buf->start = relay_alloc_buf(buf, &chan->alloc_size);
 157         if (!buf->start)
 158                 goto free_buf;
 159
 160         buf->chan = chan;
 161         kref_get(&buf->chan->kref);
 162         return buf;
 163
 164 free_buf:
 165         kfree(buf->padding);
 166         kfree(buf);
 167         return NULL;
 168 }
 169
 170 /**
 171  *      relay_destroy_channel - free the channel struct
 172  *      @kref: target kernel reference that contains the relay channel
 173  *
 174  *      Should only be called from kref_put().
 175  */
 176 void relay_destroy_channel(struct kref *kref)
 177 {
 178         struct rchan *chan = container_of(kref, struct rchan, kref);
 179         kfree(chan);
 180 }
 181
 182 /**
 183  *      relay_destroy_buf - destroy an rchan_buf struct and associated buffer
 184  *      @buf: the buffer struct
 185  */
 186 void relay_destroy_buf(struct rchan_buf *buf)
 187 {
 188         struct rchan *chan = buf->chan;
 189         unsigned int i;
 190
 191         if (likely(buf->start)) {
 192                 vunmap(buf->start);
 193                 for (i = 0; i < buf->page_count; i++)
 194                         __free_page(buf->page_array[i]);
 195                 kfree(buf->page_array);
 196         }
 197         chan->buf[buf->cpu] = NULL;
 198         kfree(buf->padding);
 199         kfree(buf);
 200         kref_put(&chan->kref, relay_destroy_channel);
 201 }
 202
 203 /**
 204  *      relay_remove_buf - remove a channel buffer
 205  *      @kref: target kernel reference that contains the relay buffer
 206  *
 207  *      Removes the file from the fileystem, which also frees the
 208  *      rchan_buf_struct and the channel buffer.  Should only be called from
 209  *      kref_put().
 210  */
 211 void relay_remove_buf(struct kref *kref)
 212 {
 213         struct rchan_buf *buf = container_of(kref, struct rchan_buf, kref);
 214         buf->chan->cb->remove_buf_file(buf->dentry);
 215         relay_destroy_buf(buf);
 216 }
 217
 218 /**
 219  *      relay_buf_empty - boolean, is the channel buffer empty?
 220  *      @buf: channel buffer
 221  *
 222  *      Returns 1 if the buffer is empty, 0 otherwise.
 223  */
 224 int relay_buf_empty(struct rchan_buf *buf)
 225 {
 226         return (buf->subbufs_produced - buf->subbufs_consumed) ? 0 : 1;
 227 }
 228 EXPORT_SYMBOL_GPL(relay_buf_empty);
 229
 230 /**
 231  *      relay_buf_full - boolean, is the channel buffer full?
 232  *      @buf: channel buffer
 233  *
 234  *      Returns 1 if the buffer is full, 0 otherwise.
 235  */
 236 int relay_buf_full(struct rchan_buf *buf)
 237 {
 238         size_t ready = buf->subbufs_produced - buf->subbufs_consumed;
 239         return (ready >= buf->chan->n_subbufs) ? 1 : 0;
 240 }
 241 EXPORT_SYMBOL_GPL(relay_buf_full);
 242
 243 /*
 244  * High-level relay kernel API and associated functions.
 245  */
 246
 247 /*
 248  * rchan_callback implementations defining default channel behavior.  Used
 249  * in place of corresponding NULL values in client callback struct.
 250  */
 251
 252 /*
 253  * subbuf_start() default callback.  Does nothing.
 254  */
 255 static int subbuf_start_default_callback (struct rchan_buf *buf,
 256                                           void *subbuf,
 257                                           void *prev_subbuf,
 258                                           size_t prev_padding)
 259 {
 260         if (relay_buf_full(buf))
 261                 return 0;
 262
 263         return 1;
 264 }
 265
 266 /*
 267  * buf_mapped() default callback.  Does nothing.
 268  */
 269 static void buf_mapped_default_callback(struct rchan_buf *buf,
 270                                         struct file *filp)
 271 {
 272 }
 273
 274 /*
 275  * buf_unmapped() default callback.  Does nothing.
 276  */
 277 static void buf_unmapped_default_callback(struct rchan_buf *buf,
 278                                           struct file *filp)
 279 {
 280 }
 281
 282 /*
 283  * create_buf_file_create() default callback.  Does nothing.
 284  */
 285 static struct dentry *create_buf_file_default_callback(const char *filename,
 286                                                        struct dentry *parent,
 287                                                        int mode,
 288                                                        struct rchan_buf *buf,
 289                                                        int *is_global)
 290 {
 291         return NULL;
 292 }
 293
 294 /*
 295  * remove_buf_file() default callback.  Does nothing.
 296  */
 297 static int remove_buf_file_default_callback(struct dentry *dentry)
 298 {
 299         return -EINVAL;
 300 }
 301
 302 /* relay channel default callbacks */
 303 static struct rchan_callbacks default_channel_callbacks = {
 304         .subbuf_start = subbuf_start_default_callback,
 305         .buf_mapped = buf_mapped_default_callback,
 306         .buf_unmapped = buf_unmapped_default_callback,
 307         .create_buf_file = create_buf_file_default_callback,
 308         .remove_buf_file = remove_buf_file_default_callback,
 309 };
 310
 311 /**
 312  *      wakeup_readers - wake up readers waiting on a channel
 313  *      @data: contains the the channel buffer
 314  *
 315  *      This is the timer function used to defer reader waking.
 316  */
 317 static void wakeup_readers(unsigned long data)
 318 {
 319         struct rchan_buf *buf = (struct rchan_buf *)data;
 320         wake_up_interruptible(&buf->read_wait);
 321 }
 322
 323 /**
 324  *      __relay_reset - reset a channel buffer
 325  *      @buf: the channel buffer
 326  *      @init: 1 if this is a first-time initialization
 327  *
 328  *      See relay_reset() for description of effect.
 329  */
 330 static void __relay_reset(struct rchan_buf *buf, unsigned int init)
 331 {
 332         size_t i;
 333
 334         if (init) {
 335                 init_waitqueue_head(&buf->read_wait);
 336                 kref_init(&buf->kref);
 337                 setup_timer(&buf->timer, wakeup_readers, (unsigned long)buf);
 338         } else
 339                 del_timer_sync(&buf->timer);
 340
 341         buf->subbufs_produced = 0;
 342         buf->subbufs_consumed = 0;
 343         buf->bytes_consumed = 0;
 344         buf->finalized = 0;
 345         buf->data = buf->start;
 346         buf->offset = 0;
 347
 348         for (i = 0; i < buf->chan->n_subbufs; i++)
 349                 buf->padding[i] = 0;
 350
 351         buf->chan->cb->subbuf_start(buf, buf->data, NULL, 0);
 352 }
 353
 354 /**
 355  *      relay_reset - reset the channel
 356  *      @chan: the channel
 357  *
 358  *      This has the effect of erasing all data from all channel buffers
 359  *      and restarting the channel in its initial state.  The buffers
 360  *      are not freed, so any mappings are still in effect.
 361  *
 362  *      NOTE. Care should be taken that the channel isn't actually
 363  *      being used by anything when this call is made.
 364  */
 365 void relay_reset(struct rchan *chan)
 366 {
 367         unsigned int i;
 368
 369         if (!chan)
 370                 return;
 371
 372         if (chan->is_global && chan->buf[0]) {
 373                 __relay_reset(chan->buf[0], 0);
 374                 return;
 375         }
 376
 377         mutex_lock(&relay_channels_mutex);
 378         for_each_online_cpu(i)
 379                 if (chan->buf[i])
 380                         __relay_reset(chan->buf[i], 0);
 381         mutex_unlock(&relay_channels_mutex);
 382 }
 383 EXPORT_SYMBOL_GPL(relay_reset);
 384
 385 /*
 386  *      relay_open_buf - create a new relay channel buffer
 387  *
 388  *      used by relay_open() and CPU hotplug.
 389  */
 390 static struct rchan_buf *relay_open_buf(struct rchan *chan, unsigned int cpu)
 391 {
 392         struct rchan_buf *buf = NULL;
 393         struct dentry *dentry;
 394         char *tmpname;
 395
 396         if (chan->is_global)
 397                 return chan->buf[0];
 398
 399         tmpname = kzalloc(NAME_MAX + 1, GFP_KERNEL);
 400         if (!tmpname)
 401                 goto end;
 402         snprintf(tmpname, NAME_MAX, "%s%d", chan->base_filename, cpu);
 403
 404         buf = relay_create_buf(chan);
 405         if (!buf)
 406                 goto free_name;
 407
 408         buf->cpu = cpu;
 409         __relay_reset(buf, 1);
 410
 411         /* Create file in fs */
 412         dentry = chan->cb->create_buf_file(tmpname, chan->parent, S_IRUSR,
 413                                            buf, &chan->is_global);
 414         if (!dentry)
 415                 goto free_buf;
 416
 417         buf->dentry = dentry;
 418
 419         if(chan->is_global) {
 420                 chan->buf[0] = buf;
 421                 buf->cpu = 0;
 422         }
 423
 424         goto free_name;
 425
 426 free_buf:
 427         relay_destroy_buf(buf);
 428 free_name:
 429         kfree(tmpname);
 430 end:
 431         return buf;
 432 }
 433
 434 /**
 435  *      relay_close_buf - close a channel buffer
 436  *      @buf: channel buffer
 437  *
 438  *      Marks the buffer finalized and restores the default callbacks.
 439  *      The channel buffer and channel buffer data structure are then freed
 440  *      automatically when the last reference is given up.
 441  */
 442 static void relay_close_buf(struct rchan_buf *buf)
 443 {
 444         buf->finalized = 1;
 445         del_timer_sync(&buf->timer);
 446         kref_put(&buf->kref, relay_remove_buf);
 447 }
 448
 449 static void setup_callbacks(struct rchan *chan,
 450                                    struct rchan_callbacks *cb)
 451 {
 452         if (!cb) {
 453                 chan->cb = &default_channel_callbacks;
 454                 return;
 455         }
 456
 457         if (!cb->subbuf_start)
 458                 cb->subbuf_start = subbuf_start_default_callback;
 459         if (!cb->buf_mapped)
 460                 cb->buf_mapped = buf_mapped_default_callback;
 461         if (!cb->buf_unmapped)
 462                 cb->buf_unmapped = buf_unmapped_default_callback;
 463         if (!cb->create_buf_file)
 464                 cb->create_buf_file = create_buf_file_default_callback;
 465         if (!cb->remove_buf_file)
 466                 cb->remove_buf_file = remove_buf_file_default_callback;
 467         chan->cb = cb;
 468 }
 469
 470 /**
 471  *      relay_hotcpu_callback - CPU hotplug callback
 472  *      @nb: notifier block
 473  *      @action: hotplug action to take
 474  *      @hcpu: CPU number
 475  *
 476  *      Returns the success/failure of the operation. (%NOTIFY_OK, %NOTIFY_BAD)
 477  */
 478 static int __cpuinit relay_hotcpu_callback(struct notifier_block *nb,
 479                                 unsigned long action,
 480                                 void *hcpu)
 481 {
 482         unsigned int hotcpu = (unsigned long)hcpu;
 483         struct rchan *chan;
 484
 485         switch(action) {
 486         case CPU_UP_PREPARE:
 487                 mutex_lock(&relay_channels_mutex);
 488                 list_for_each_entry(chan, &relay_channels, list) {
 489                         if (chan->buf[hotcpu])
 490                                 continue;
 491                         chan->buf[hotcpu] = relay_open_buf(chan, hotcpu);
 492                         if(!chan->buf[hotcpu]) {
 493                                 printk(KERN_ERR
 494                                         "relay_hotcpu_callback: cpu %d buffer "
 495                                         "creation failed\n", hotcpu);
 496                                 mutex_unlock(&relay_channels_mutex);
 497                                 return NOTIFY_BAD;
 498                         }
 499                 }
 500                 mutex_unlock(&relay_channels_mutex);
 501                 break;
 502         case CPU_DEAD:
 503                 /* No need to flush the cpu : will be flushed upon
 504                  * final relay_flush() call. */
 505                 break;
 506         }
 507         return NOTIFY_OK;
 508 }
 509
 510 /**
 511  *      relay_open - create a new relay channel
 512  *      @base_filename: base name of files to create
 513  *      @parent: dentry of parent directory, %NULL for root directory
 514  *      @subbuf_size: size of sub-buffers
 515  *      @n_subbufs: number of sub-buffers
 516  *      @cb: client callback functions
 517  *      @private_data: user-defined data
 518  *
 519  *      Returns channel pointer if successful, %NULL otherwise.
 520  *
 521  *      Creates a channel buffer for each cpu using the sizes and
 522  *      attributes specified.  The created channel buffer files
 523  *      will be named base_filename0...base_filenameN-1.  File
 524  *      permissions will be %S_IRUSR.
 525  */
 526 struct rchan *relay_open(const char *base_filename,
 527                          struct dentry *parent,
 528                          size_t subbuf_size,
 529                          size_t n_subbufs,
 530                          struct rchan_callbacks *cb,
 531                          void *private_data)
 532 {
 533         unsigned int i;
 534         struct rchan *chan;
 535         if (!base_filename)
 536                 return NULL;
 537
 538         if (!(subbuf_size && n_subbufs))
 539                 return NULL;
 540
 541         chan = kzalloc(sizeof(struct rchan), GFP_KERNEL);
 542         if (!chan)
 543                 return NULL;
 544
 545         chan->version = RELAYFS_CHANNEL_VERSION;
 546         chan->n_subbufs = n_subbufs;
 547         chan->subbuf_size = subbuf_size;
 548         chan->alloc_size = FIX_SIZE(subbuf_size * n_subbufs);
 549         chan->parent = parent;
 550         chan->private_data = private_data;
 551         strlcpy(chan->base_filename, base_filename, NAME_MAX);
 552         setup_callbacks(chan, cb);
 553         kref_init(&chan->kref);
 554
 555         mutex_lock(&relay_channels_mutex);
 556         for_each_online_cpu(i) {
 557                 chan->buf[i] = relay_open_buf(chan, i);
 558                 if (!chan->buf[i])
 559                         goto free_bufs;
 560         }
 561         list_add(&chan->list, &relay_channels);
 562         mutex_unlock(&relay_channels_mutex);
 563
 564         return chan;
 565
 566 free_bufs:
 567         for_each_online_cpu(i) {
 568                 if (!chan->buf[i])
 569                         break;
 570                 relay_close_buf(chan->buf[i]);
 571         }
 572
 573         kref_put(&chan->kref, relay_destroy_channel);
 574         mutex_unlock(&relay_channels_mutex);
 575         return NULL;
 576 }
 577 EXPORT_SYMBOL_GPL(relay_open);
 578
 579 /**
 580  *      relay_switch_subbuf - switch to a new sub-buffer
 581  *      @buf: channel buffer
 582  *      @length: size of current event
 583  *
 584  *      Returns either the length passed in or 0 if full.
 585  *
 586  *      Performs sub-buffer-switch tasks such as invoking callbacks,
 587  *      updating padding counts, waking up readers, etc.
 588  */
 589 size_t relay_switch_subbuf(struct rchan_buf *buf, size_t length)
 590 {
 591         void *old, *new;
 592         size_t old_subbuf, new_subbuf;
 593
 594         if (unlikely(length > buf->chan->subbuf_size))
 595                 goto toobig;
 596
 597         if (buf->offset != buf->chan->subbuf_size + 1) {
 598                 buf->prev_padding = buf->chan->subbuf_size - buf->offset;
 599                 old_subbuf = buf->subbufs_produced % buf->chan->n_subbufs;
 600                 buf->padding[old_subbuf] = buf->prev_padding;
 601                 buf->subbufs_produced++;
 602                 buf->dentry->d_inode->i_size += buf->chan->subbuf_size -
 603                         buf->padding[old_subbuf];
 604                 smp_mb();
 605                 if (waitqueue_active(&buf->read_wait))
 606                         /*
 607                          * Calling wake_up_interruptible() from here
 608                          * will deadlock if we happen to be logging
 609                          * from the scheduler (trying to re-grab
 610                          * rq->lock), so defer it.
 611                          */
 612                         __mod_timer(&buf->timer, jiffies + 1);
 613         }
 614
 615         old = buf->data;
 616         new_subbuf = buf->subbufs_produced % buf->chan->n_subbufs;
 617         new = buf->start + new_subbuf * buf->chan->subbuf_size;
 618         buf->offset = 0;
 619         if (!buf->chan->cb->subbuf_start(buf, new, old, buf->prev_padding)) {
 620                 buf->offset = buf->chan->subbuf_size + 1;
 621                 return 0;
 622         }
 623         buf->data = new;
 624         buf->padding[new_subbuf] = 0;
 625
 626         if (unlikely(length + buf->offset > buf->chan->subbuf_size))
 627                 goto toobig;
 628
 629         return length;
 630
 631 toobig:
 632         buf->chan->last_toobig = length;
 633         return 0;
 634 }
 635 EXPORT_SYMBOL_GPL(relay_switch_subbuf);
 636
 637 /**
 638  *      relay_subbufs_consumed - update the buffer's sub-buffers-consumed count
 639  *      @chan: the channel
 640  *      @cpu: the cpu associated with the channel buffer to update
 641  *      @subbufs_consumed: number of sub-buffers to add to current buf's count
 642  *
 643  *      Adds to the channel buffer's consumed sub-buffer count.
 644  *      subbufs_consumed should be the number of sub-buffers newly consumed,
 645  *      not the total consumed.
 646  *
 647  *      NOTE. Kernel clients don't need to call this function if the channel
 648  *      mode is 'overwrite'.
 649  */
 650 void relay_subbufs_consumed(struct rchan *chan,
 651                             unsigned int cpu,
 652                             size_t subbufs_consumed)
 653 {
 654         struct rchan_buf *buf;
 655
 656         if (!chan)
 657                 return;
 658
 659         if (cpu >= NR_CPUS || !chan->buf[cpu])
 660                 return;
 661
 662         buf = chan->buf[cpu];
 663         buf->subbufs_consumed += subbufs_consumed;
 664         if (buf->subbufs_consumed > buf->subbufs_produced)
 665                 buf->subbufs_consumed = buf->subbufs_produced;
 666 }
 667 EXPORT_SYMBOL_GPL(relay_subbufs_consumed);
 668
 669 /**
 670  *      relay_close - close the channel
 671  *      @chan: the channel
 672  *
 673  *      Closes all channel buffers and frees the channel.
 674  */
 675 void relay_close(struct rchan *chan)
 676 {
 677         unsigned int i;
 678
 679         if (!chan)
 680                 return;
 681
 682         mutex_lock(&relay_channels_mutex);
 683         if (chan->is_global && chan->buf[0])
 684                 relay_close_buf(chan->buf[0]);
 685         else
 686                 for_each_possible_cpu(i)
 687                         if (chan->buf[i])
 688                                 relay_close_buf(chan->buf[i]);
 689
 690         if (chan->last_toobig)
 691                 printk(KERN_WARNING "relay: one or more items not logged "
 692                        "[item size (%Zd) > sub-buffer size (%Zd)]\n",
 693                        chan->last_toobig, chan->subbuf_size);
 694
 695         list_del(&chan->list);
 696         kref_put(&chan->kref, relay_destroy_channel);
 697         mutex_unlock(&relay_channels_mutex);
 698 }
 699 EXPORT_SYMBOL_GPL(relay_close);
 700
 701 /**
 702  *      relay_flush - close the channel
 703  *      @chan: the channel
 704  *
 705  *      Flushes all channel buffers, i.e. forces buffer switch.
 706  */
 707 void relay_flush(struct rchan *chan)
 708 {
 709         unsigned int i;
 710
 711         if (!chan)
 712                 return;
 713
 714         if (chan->is_global && chan->buf[0]) {
 715                 relay_switch_subbuf(chan->buf[0], 0);
 716                 return;
 717         }
 718
 719         mutex_lock(&relay_channels_mutex);
 720         for_each_possible_cpu(i)
 721                 if (chan->buf[i])
 722                         relay_switch_subbuf(chan->buf[i], 0);
 723         mutex_unlock(&relay_channels_mutex);
 724 }
 725 EXPORT_SYMBOL_GPL(relay_flush);
 726
 727 /**
 728  *      relay_file_open - open file op for relay files
 729  *      @inode: the inode
 730  *      @filp: the file
 731  *
 732  *      Increments the channel buffer refcount.
 733  */
 734 static int relay_file_open(struct inode *inode, struct file *filp)
 735 {
 736         struct rchan_buf *buf = inode->i_private;
 737         kref_get(&buf->kref);
 738         filp->private_data = buf;
 739
 740         return 0;
 741 }
 742
 743 /**
 744  *      relay_file_mmap - mmap file op for relay files
 745  *      @filp: the file
 746  *      @vma: the vma describing what to map
 747  *
 748  *      Calls upon relay_mmap_buf() to map the file into user space.
 749  */
 750 static int relay_file_mmap(struct file *filp, struct vm_area_struct *vma)
 751 {
 752         struct rchan_buf *buf = filp->private_data;
 753         return relay_mmap_buf(buf, vma);
 754 }
 755
 756 /**
 757  *      relay_file_poll - poll file op for relay files
 758  *      @filp: the file
 759  *      @wait: poll table
 760  *
 761  *      Poll implemention.
 762  */
 763 static unsigned int relay_file_poll(struct file *filp, poll_table *wait)
 764 {
 765         unsigned int mask = 0;
 766         struct rchan_buf *buf = filp->private_data;
 767
 768         if (buf->finalized)
 769                 return POLLERR;
 770
 771         if (filp->f_mode & FMODE_READ) {
 772                 poll_wait(filp, &buf->read_wait, wait);
 773                 if (!relay_buf_empty(buf))
 774                         mask |= POLLIN | POLLRDNORM;
 775         }
 776
 777         return mask;
 778 }
 779
 780 /**
 781  *      relay_file_release - release file op for relay files
 782  *      @inode: the inode
 783  *      @filp: the file
 784  *
 785  *      Decrements the channel refcount, as the filesystem is
 786  *      no longer using it.
 787  */
 788 static int relay_file_release(struct inode *inode, struct file *filp)
 789 {
 790         struct rchan_buf *buf = filp->private_data;
 791         kref_put(&buf->kref, relay_remove_buf);
 792
 793         return 0;
 794 }
 795
 796 /*
 797  *      relay_file_read_consume - update the consumed count for the buffer
 798  */
 799 static void relay_file_read_consume(struct rchan_buf *buf,
 800                                     size_t read_pos,
 801                                     size_t bytes_consumed)
 802 {
 803         size_t subbuf_size = buf->chan->subbuf_size;
 804         size_t n_subbufs = buf->chan->n_subbufs;
 805         size_t read_subbuf;
 806
 807         if (buf->bytes_consumed + bytes_consumed > subbuf_size) {
 808                 relay_subbufs_consumed(buf->chan, buf->cpu, 1);
 809                 buf->bytes_consumed = 0;
 810         }
 811
 812         buf->bytes_consumed += bytes_consumed;
 813         read_subbuf = read_pos / buf->chan->subbuf_size;
 814         if (buf->bytes_consumed + buf->padding[read_subbuf] == subbuf_size) {
 815                 if ((read_subbuf == buf->subbufs_produced % n_subbufs) &&
 816                     (buf->offset == subbuf_size))
 817                         return;
 818                 relay_subbufs_consumed(buf->chan, buf->cpu, 1);
 819                 buf->bytes_consumed = 0;
 820         }
 821 }
 822
 823 /*
 824  *      relay_file_read_avail - boolean, are there unconsumed bytes available?
 825  */
 826 static int relay_file_read_avail(struct rchan_buf *buf, size_t read_pos)
 827 {
 828         size_t subbuf_size = buf->chan->subbuf_size;
 829         size_t n_subbufs = buf->chan->n_subbufs;
 830         size_t produced = buf->subbufs_produced;
 831         size_t consumed = buf->subbufs_consumed;
 832
 833         relay_file_read_consume(buf, read_pos, 0);
 834
 835         if (unlikely(buf->offset > subbuf_size)) {
 836                 if (produced == consumed)
 837                         return 0;
 838                 return 1;
 839         }
 840
 841         if (unlikely(produced - consumed >= n_subbufs)) {
 842                 consumed = (produced / n_subbufs) * n_subbufs;
 843                 buf->subbufs_consumed = consumed;
 844         }
 845
 846         produced = (produced % n_subbufs) * subbuf_size + buf->offset;
 847         consumed = (consumed % n_subbufs) * subbuf_size + buf->bytes_consumed;
 848
 849         if (consumed > produced)
 850                 produced += n_subbufs * subbuf_size;
 851
 852         if (consumed == produced)
 853                 return 0;
 854
 855         return 1;
 856 }
 857
 858 /**
 859  *      relay_file_read_subbuf_avail - return bytes available in sub-buffer
 860  *      @read_pos: file read position
 861  *      @buf: relay channel buffer
 862  */
 863 static size_t relay_file_read_subbuf_avail(size_t read_pos,
 864                                            struct rchan_buf *buf)
 865 {
 866         size_t padding, avail = 0;
 867         size_t read_subbuf, read_offset, write_subbuf, write_offset;
 868         size_t subbuf_size = buf->chan->subbuf_size;
 869
 870         write_subbuf = (buf->data - buf->start) / subbuf_size;
 871         write_offset = buf->offset > subbuf_size ? subbuf_size : buf->offset;
 872         read_subbuf = read_pos / subbuf_size;
 873         read_offset = read_pos % subbuf_size;
 874         padding = buf->padding[read_subbuf];
 875
 876         if (read_subbuf == write_subbuf) {
 877                 if (read_offset + padding < write_offset)
 878                         avail = write_offset - (read_offset + padding);
 879         } else
 880                 avail = (subbuf_size - padding) - read_offset;
 881
 882         return avail;
 883 }
 884
 885 /**
 886  *      relay_file_read_start_pos - find the first available byte to read
 887  *      @read_pos: file read position
 888  *      @buf: relay channel buffer
 889  *
 890  *      If the @read_pos is in the middle of padding, return the
 891  *      position of the first actually available byte, otherwise
 892  *      return the original value.
 893  */
 894 static size_t relay_file_read_start_pos(size_t read_pos,
 895                                         struct rchan_buf *buf)
 896 {
 897         size_t read_subbuf, padding, padding_start, padding_end;
 898         size_t subbuf_size = buf->chan->subbuf_size;
 899         size_t n_subbufs = buf->chan->n_subbufs;
 900
 901         read_subbuf = read_pos / subbuf_size;
 902         padding = buf->padding[read_subbuf];
 903         padding_start = (read_subbuf + 1) * subbuf_size - padding;
 904         padding_end = (read_subbuf + 1) * subbuf_size;
 905         if (read_pos >= padding_start && read_pos < padding_end) {
 906                 read_subbuf = (read_subbuf + 1) % n_subbufs;
 907                 read_pos = read_subbuf * subbuf_size;
 908         }
 909
 910         return read_pos;
 911 }
 912
 913 /**
 914  *      relay_file_read_end_pos - return the new read position
 915  *      @read_pos: file read position
 916  *      @buf: relay channel buffer
 917  *      @count: number of bytes to be read
 918  */
 919 static size_t relay_file_read_end_pos(struct rchan_buf *buf,
 920                                       size_t read_pos,
 921                                       size_t count)
 922 {
 923         size_t read_subbuf, padding, end_pos;
 924         size_t subbuf_size = buf->chan->subbuf_size;
 925         size_t n_subbufs = buf->chan->n_subbufs;
 926
 927         read_subbuf = read_pos / subbuf_size;
 928         padding = buf->padding[read_subbuf];
 929         if (read_pos % subbuf_size + count + padding == subbuf_size)
 930                 end_pos = (read_subbuf + 1) * subbuf_size;
 931         else
 932                 end_pos = read_pos + count;
 933         if (end_pos >= subbuf_size * n_subbufs)
 934                 end_pos = 0;
 935
 936         return end_pos;
 937 }
 938
 939 /*
 940  *      subbuf_read_actor - read up to one subbuf's worth of data
 941  */
 942 static int subbuf_read_actor(size_t read_start,
 943                              struct rchan_buf *buf,
 944                              size_t avail,
 945                              read_descriptor_t *desc,
 946                              read_actor_t actor)
 947 {
 948         void *from;
 949         int ret = 0;
 950
 951         from = buf->start + read_start;
 952         ret = avail;
 953         if (copy_to_user(desc->arg.buf, from, avail)) {
 954                 desc->error = -EFAULT;
 955                 ret = 0;
 956         }
 957         desc->arg.data += ret;
 958         desc->written += ret;
 959         desc->count -= ret;
 960
 961         return ret;
 962 }
 963
 964 /*
 965  *      subbuf_send_actor - send up to one subbuf's worth of data
 966  */
 967 static int subbuf_send_actor(size_t read_start,
 968                              struct rchan_buf *buf,
 969                              size_t avail,
 970                              read_descriptor_t *desc,
 971                              read_actor_t actor)
 972 {
 973         unsigned long pidx, poff;
 974         unsigned int subbuf_pages;
 975         int ret = 0;
 976
 977         subbuf_pages = buf->chan->alloc_size >> PAGE_SHIFT;
 978         pidx = (read_start / PAGE_SIZE) % subbuf_pages;
 979         poff = read_start & ~PAGE_MASK;
 980         while (avail) {
 981                 struct page *p = buf->page_array[pidx];
 982                 unsigned int len;
 983
 984                 len = PAGE_SIZE - poff;
 985                 if (len > avail)
 986                         len = avail;
 987
 988                 len = actor(desc, p, poff, len);
 989                 if (desc->error)
 990                         break;
 991
 992                 avail -= len;
 993                 ret += len;
 994                 poff = 0;
 995                 pidx = (pidx + 1) % subbuf_pages;
 996         }
 997
 998         return ret;
 999 }
1000
1001 typedef int (*subbuf_actor_t) (size_t read_start,
1002                                struct rchan_buf *buf,
1003                                size_t avail,
1004                                read_descriptor_t *desc,
1005                                read_actor_t actor);
1006
1007 /*
1008  *      relay_file_read_subbufs - read count bytes, bridging subbuf boundaries
1009  */
1010 static ssize_t relay_file_read_subbufs(struct file *filp, loff_t *ppos,
1011                                         subbuf_actor_t subbuf_actor,
1012                                         read_actor_t actor,
1013                                         read_descriptor_t *desc)
1014 {
1015         struct rchan_buf *buf = filp->private_data;
1016         size_t read_start, avail;
1017         int ret;
1018
1019         if (!desc->count)
1020                 return 0;
1021
1022         mutex_lock(&filp->f_path.dentry->d_inode->i_mutex);
1023         do {
1024                 if (!relay_file_read_avail(buf, *ppos))
1025                         break;
1026
1027                 read_start = relay_file_read_start_pos(*ppos, buf);
1028                 avail = relay_file_read_subbuf_avail(read_start, buf);
1029                 if (!avail)
1030                         break;
1031
1032                 avail = min(desc->count, avail);
1033                 ret = subbuf_actor(read_start, buf, avail, desc, actor);
1034                 if (desc->error < 0)
1035                         break;
1036
1037                 if (ret) {
1038                         relay_file_read_consume(buf, read_start, ret);
1039                         *ppos = relay_file_read_end_pos(buf, read_start, ret);
1040                 }
1041         } while (desc->count && ret);
1042         mutex_unlock(&filp->f_path.dentry->d_inode->i_mutex);
1043
1044         return desc->written;
1045 }
1046
1047 static ssize_t relay_file_read(struct file *filp,
1048                                char __user *buffer,
1049                                size_t count,
1050                                loff_t *ppos)
1051 {
1052         read_descriptor_t desc;
1053         desc.written = 0;
1054         desc.count = count;
1055         desc.arg.buf = buffer;
1056         desc.error = 0;
1057         return relay_file_read_subbufs(filp, ppos, subbuf_read_actor,
1058                                        NULL, &desc);
1059 }
1060
1061 static ssize_t relay_file_sendfile(struct file *filp,
1062                                    loff_t *ppos,
1063                                    size_t count,
1064                                    read_actor_t actor,
1065                                    void *target)
1066 {
1067         read_descriptor_t desc;
1068         desc.written = 0;
1069         desc.count = count;
1070         desc.arg.data = target;
1071         desc.error = 0;
1072         return relay_file_read_subbufs(filp, ppos, subbuf_send_actor,
1073                                        actor, &desc);
1074 }
1075
1076 const struct file_operations relay_file_operations = {
1077         .open           = relay_file_open,
1078         .poll           = relay_file_poll,
1079         .mmap           = relay_file_mmap,
1080         .read           = relay_file_read,
1081         .llseek         = no_llseek,
1082         .release        = relay_file_release,
1083         .sendfile       = relay_file_sendfile,
1084 };
1085 EXPORT_SYMBOL_GPL(relay_file_operations);
1086
1087 static __init int relay_init(void)
1088 {
1089
1090         hotcpu_notifier(relay_hotcpu_callback, 0);
1091         return 0;
1092 }
1093
1094 module_init(relay_init);