kernel/relay.c

   1 /*
   2  * Public API and common code for kernel->userspace relay file support.
   3  *
   4  * See Documentation/filesystems/relay.txt for an overview.
   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 #include <linux/splice.h>
  25
  26 /* list of open channels, for cpu hotplug */
  27 static DEFINE_MUTEX(relay_channels_mutex);
  28 static LIST_HEAD(relay_channels);
  29
  30 /*
  31  * close() vm_op implementation for relay file mapping.
  32  */
  33 static void relay_file_mmap_close(struct vm_area_struct *vma)
  34 {
  35         struct rchan_buf *buf = vma->vm_private_data;
  36         buf->chan->cb->buf_unmapped(buf, vma->vm_file);
  37 }
  38
  39 /*
  40  * fault() vm_op implementation for relay file mapping.
  41  */
  42 static int relay_buf_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
  43 {
  44         struct page *page;
  45         struct rchan_buf *buf = vma->vm_private_data;
  46         pgoff_t pgoff = vmf->pgoff;
  47
  48         if (!buf)
  49                 return VM_FAULT_OOM;
  50
  51         page = vmalloc_to_page(buf->start + (pgoff << PAGE_SHIFT));
  52         if (!page)
  53                 return VM_FAULT_SIGBUS;
  54         get_page(page);
  55         vmf->page = page;
  56
  57         return 0;
  58 }
  59
  60 /*
  61  * vm_ops for relay file mappings.
  62  */
  63 static struct vm_operations_struct relay_file_mmap_ops = {
  64         .fault = relay_buf_fault,
  65         .close = relay_file_mmap_close,
  66 };
  67
  68 /*
  69  * allocate an array of pointers of struct page
  70  */
  71 static struct page **relay_alloc_page_array(unsigned int n_pages)
  72 {
  73         struct page **array;
  74         size_t pa_size = n_pages * sizeof(struct page *);
  75
  76         if (pa_size > PAGE_SIZE) {
  77                 array = vmalloc(pa_size);
  78                 if (array)
  79                         memset(array, 0, pa_size);
  80         } else {
  81                 array = kzalloc(pa_size, GFP_KERNEL);
  82         }
  83         return array;
  84 }
  85
  86 /*
  87  * free an array of pointers of struct page
  88  */
  89 static void relay_free_page_array(struct page **array)
  90 {
  91         if (is_vmalloc_addr(array))
  92                 vfree(array);
  93         else
  94                 kfree(array);
  95 }
  96
  97 /**
  98  *      relay_mmap_buf: - mmap channel buffer to process address space
  99  *      @buf: relay channel buffer
 100  *      @vma: vm_area_struct describing memory to be mapped
 101  *
 102  *      Returns 0 if ok, negative on error
 103  *
 104  *      Caller should already have grabbed mmap_sem.
 105  */
 106 static int relay_mmap_buf(struct rchan_buf *buf, struct vm_area_struct *vma)
 107 {
 108         unsigned long length = vma->vm_end - vma->vm_start;
 109         struct file *filp = vma->vm_file;
 110
 111         if (!buf)
 112                 return -EBADF;
 113
 114         if (length != (unsigned long)buf->chan->alloc_size)
 115                 return -EINVAL;
 116
 117         vma->vm_ops = &relay_file_mmap_ops;
 118         vma->vm_flags |= VM_DONTEXPAND;
 119         vma->vm_private_data = buf;
 120         buf->chan->cb->buf_mapped(buf, filp);
 121
 122         return 0;
 123 }
 124
 125 /**
 126  *      relay_alloc_buf - allocate a channel buffer
 127  *      @buf: the buffer struct
 128  *      @size: total size of the buffer
 129  *
 130  *      Returns a pointer to the resulting buffer, %NULL if unsuccessful. The
 131  *      passed in size will get page aligned, if it isn't already.
 132  */
 133 static void *relay_alloc_buf(struct rchan_buf *buf, size_t *size)
 134 {
 135         void *mem;
 136         unsigned int i, j, n_pages;
 137
 138         *size = PAGE_ALIGN(*size);
 139         n_pages = *size >> PAGE_SHIFT;
 140
 141         buf->page_array = relay_alloc_page_array(n_pages);
 142         if (!buf->page_array)
 143                 return NULL;
 144
 145         for (i = 0; i < n_pages; i++) {
 146                 buf->page_array[i] = alloc_page(GFP_KERNEL);
 147                 if (unlikely(!buf->page_array[i]))
 148                         goto depopulate;
 149                 set_page_private(buf->page_array[i], (unsigned long)buf);
 150         }
 151         mem = vmap(buf->page_array, n_pages, VM_MAP, PAGE_KERNEL);
 152         if (!mem)
 153                 goto depopulate;
 154
 155         memset(mem, 0, *size);
 156         buf->page_count = n_pages;
 157         return mem;
 158
 159 depopulate:
 160         for (j = 0; j < i; j++)
 161                 __free_page(buf->page_array[j]);
 162         relay_free_page_array(buf->page_array);
 163         return NULL;
 164 }
 165
 166 /**
 167  *      relay_create_buf - allocate and initialize a channel buffer
 168  *      @chan: the relay channel
 169  *
 170  *      Returns channel buffer if successful, %NULL otherwise.
 171  */
 172 static struct rchan_buf *relay_create_buf(struct rchan *chan)
 173 {
 174         struct rchan_buf *buf = kzalloc(sizeof(struct rchan_buf), GFP_KERNEL);
 175         if (!buf)
 176                 return NULL;
 177
 178         buf->padding = kmalloc(chan->n_subbufs * sizeof(size_t *), GFP_KERNEL);
 179         if (!buf->padding)
 180                 goto free_buf;
 181
 182         buf->start = relay_alloc_buf(buf, &chan->alloc_size);
 183         if (!buf->start)
 184                 goto free_buf;
 185
 186         buf->chan = chan;
 187         kref_get(&buf->chan->kref);
 188         return buf;
 189
 190 free_buf:
 191         kfree(buf->padding);
 192         kfree(buf);
 193         return NULL;
 194 }
 195
 196 /**
 197  *      relay_destroy_channel - free the channel struct
 198  *      @kref: target kernel reference that contains the relay channel
 199  *
 200  *      Should only be called from kref_put().
 201  */
 202 static void relay_destroy_channel(struct kref *kref)
 203 {
 204         struct rchan *chan = container_of(kref, struct rchan, kref);
 205         kfree(chan);
 206 }
 207
 208 /**
 209  *      relay_destroy_buf - destroy an rchan_buf struct and associated buffer
 210  *      @buf: the buffer struct
 211  */
 212 static void relay_destroy_buf(struct rchan_buf *buf)
 213 {
 214         struct rchan *chan = buf->chan;
 215         unsigned int i;
 216
 217         if (likely(buf->start)) {
 218                 vunmap(buf->start);
 219                 for (i = 0; i < buf->page_count; i++)
 220                         __free_page(buf->page_array[i]);
 221                 relay_free_page_array(buf->page_array);
 222         }
 223         chan->buf[buf->cpu] = NULL;
 224         kfree(buf->padding);
 225         kfree(buf);
 226         kref_put(&chan->kref, relay_destroy_channel);
 227 }
 228
 229 /**
 230  *      relay_remove_buf - remove a channel buffer
 231  *      @kref: target kernel reference that contains the relay buffer
 232  *
 233  *      Removes the file from the fileystem, which also frees the
 234  *      rchan_buf_struct and the channel buffer.  Should only be called from
 235  *      kref_put().
 236  */
 237 static void relay_remove_buf(struct kref *kref)
 238 {
 239         struct rchan_buf *buf = container_of(kref, struct rchan_buf, kref);
 240         buf->chan->cb->remove_buf_file(buf->dentry);
 241         relay_destroy_buf(buf);
 242 }
 243
 244 /**
 245  *      relay_buf_empty - boolean, is the channel buffer empty?
 246  *      @buf: channel buffer
 247  *
 248  *      Returns 1 if the buffer is empty, 0 otherwise.
 249  */
 250 static int relay_buf_empty(struct rchan_buf *buf)
 251 {
 252         return (buf->subbufs_produced - buf->subbufs_consumed) ? 0 : 1;
 253 }
 254
 255 /**
 256  *      relay_buf_full - boolean, is the channel buffer full?
 257  *      @buf: channel buffer
 258  *
 259  *      Returns 1 if the buffer is full, 0 otherwise.
 260  */
 261 int relay_buf_full(struct rchan_buf *buf)
 262 {
 263         size_t ready = buf->subbufs_produced - buf->subbufs_consumed;
 264         return (ready >= buf->chan->n_subbufs) ? 1 : 0;
 265 }
 266 EXPORT_SYMBOL_GPL(relay_buf_full);
 267
 268 /*
 269  * High-level relay kernel API and associated functions.
 270  */
 271
 272 /*
 273  * rchan_callback implementations defining default channel behavior.  Used
 274  * in place of corresponding NULL values in client callback struct.
 275  */
 276
 277 /*
 278  * subbuf_start() default callback.  Does nothing.
 279  */
 280 static int subbuf_start_default_callback (struct rchan_buf *buf,
 281                                           void *subbuf,
 282                                           void *prev_subbuf,
 283                                           size_t prev_padding)
 284 {
 285         if (relay_buf_full(buf))
 286                 return 0;
 287
 288         return 1;
 289 }
 290
 291 /*
 292  * buf_mapped() default callback.  Does nothing.
 293  */
 294 static void buf_mapped_default_callback(struct rchan_buf *buf,
 295                                         struct file *filp)
 296 {
 297 }
 298
 299 /*
 300  * buf_unmapped() default callback.  Does nothing.
 301  */
 302 static void buf_unmapped_default_callback(struct rchan_buf *buf,
 303                                           struct file *filp)
 304 {
 305 }
 306
 307 /*
 308  * create_buf_file_create() default callback.  Does nothing.
 309  */
 310 static struct dentry *create_buf_file_default_callback(const char *filename,
 311                                                        struct dentry *parent,
 312                                                        int mode,
 313                                                        struct rchan_buf *buf,
 314                                                        int *is_global)
 315 {
 316         return NULL;
 317 }
 318
 319 /*
 320  * remove_buf_file() default callback.  Does nothing.
 321  */
 322 static int remove_buf_file_default_callback(struct dentry *dentry)
 323 {
 324         return -EINVAL;
 325 }
 326
 327 /* relay channel default callbacks */
 328 static struct rchan_callbacks default_channel_callbacks = {
 329         .subbuf_start = subbuf_start_default_callback,
 330         .buf_mapped = buf_mapped_default_callback,
 331         .buf_unmapped = buf_unmapped_default_callback,
 332         .create_buf_file = create_buf_file_default_callback,
 333         .remove_buf_file = remove_buf_file_default_callback,
 334 };
 335
 336 /**
 337  *      wakeup_readers - wake up readers waiting on a channel
 338  *      @data: contains the channel buffer
 339  *
 340  *      This is the timer function used to defer reader waking.
 341  */
 342 static void wakeup_readers(unsigned long data)
 343 {
 344         struct rchan_buf *buf = (struct rchan_buf *)data;
 345         wake_up_interruptible(&buf->read_wait);
 346 }
 347
 348 /**
 349  *      __relay_reset - reset a channel buffer
 350  *      @buf: the channel buffer
 351  *      @init: 1 if this is a first-time initialization
 352  *
 353  *      See relay_reset() for description of effect.
 354  */
 355 static void __relay_reset(struct rchan_buf *buf, unsigned int init)
 356 {
 357         size_t i;
 358
 359         if (init) {
 360                 init_waitqueue_head(&buf->read_wait);
 361                 kref_init(&buf->kref);
 362                 setup_timer(&buf->timer, wakeup_readers, (unsigned long)buf);
 363         } else
 364                 del_timer_sync(&buf->timer);
 365
 366         buf->subbufs_produced = 0;
 367         buf->subbufs_consumed = 0;
 368         buf->bytes_consumed = 0;
 369         buf->finalized = 0;
 370         buf->data = buf->start;
 371         buf->offset = 0;
 372
 373         for (i = 0; i < buf->chan->n_subbufs; i++)
 374                 buf->padding[i] = 0;
 375
 376         buf->chan->cb->subbuf_start(buf, buf->data, NULL, 0);
 377 }
 378
 379 /**
 380  *      relay_reset - reset the channel
 381  *      @chan: the channel
 382  *
 383  *      This has the effect of erasing all data from all channel buffers
 384  *      and restarting the channel in its initial state.  The buffers
 385  *      are not freed, so any mappings are still in effect.
 386  *
 387  *      NOTE. Care should be taken that the channel isn't actually
 388  *      being used by anything when this call is made.
 389  */
 390 void relay_reset(struct rchan *chan)
 391 {
 392         unsigned int i;
 393
 394         if (!chan)
 395                 return;
 396
 397         if (chan->is_global && chan->buf[0]) {
 398                 __relay_reset(chan->buf[0], 0);
 399                 return;
 400         }
 401
 402         mutex_lock(&relay_channels_mutex);
 403         for_each_online_cpu(i)
 404                 if (chan->buf[i])
 405                         __relay_reset(chan->buf[i], 0);
 406         mutex_unlock(&relay_channels_mutex);
 407 }
 408 EXPORT_SYMBOL_GPL(relay_reset);
 409
 410 /*
 411  *      relay_open_buf - create a new relay channel buffer
 412  *
 413  *      used by relay_open() and CPU hotplug.
 414  */
 415 static struct rchan_buf *relay_open_buf(struct rchan *chan, unsigned int cpu)
 416 {
 417         struct rchan_buf *buf = NULL;
 418         struct dentry *dentry;
 419         char *tmpname;
 420
 421         if (chan->is_global)
 422                 return chan->buf[0];
 423
 424         tmpname = kzalloc(NAME_MAX + 1, GFP_KERNEL);
 425         if (!tmpname)
 426                 goto end;
 427         snprintf(tmpname, NAME_MAX, "%s%d", chan->base_filename, cpu);
 428
 429         buf = relay_create_buf(chan);
 430         if (!buf)
 431                 goto free_name;
 432
 433         buf->cpu = cpu;
 434         __relay_reset(buf, 1);
 435
 436         /* Create file in fs */
 437         dentry = chan->cb->create_buf_file(tmpname, chan->parent, S_IRUSR,
 438                                            buf, &chan->is_global);
 439         if (!dentry)
 440                 goto free_buf;
 441
 442         buf->dentry = dentry;
 443
 444         if(chan->is_global) {
 445                 chan->buf[0] = buf;
 446                 buf->cpu = 0;
 447         }
 448
 449         goto free_name;
 450
 451 free_buf:
 452         relay_destroy_buf(buf);
 453         buf = NULL;
 454 free_name:
 455         kfree(tmpname);
 456 end:
 457         return buf;
 458 }
 459
 460 /**
 461  *      relay_close_buf - close a channel buffer
 462  *      @buf: channel buffer
 463  *
 464  *      Marks the buffer finalized and restores the default callbacks.
 465  *      The channel buffer and channel buffer data structure are then freed
 466  *      automatically when the last reference is given up.
 467  */
 468 static void relay_close_buf(struct rchan_buf *buf)
 469 {
 470         buf->finalized = 1;
 471         del_timer_sync(&buf->timer);
 472         kref_put(&buf->kref, relay_remove_buf);
 473 }
 474
 475 static void setup_callbacks(struct rchan *chan,
 476                                    struct rchan_callbacks *cb)
 477 {
 478         if (!cb) {
 479                 chan->cb = &default_channel_callbacks;
 480                 return;
 481         }
 482
 483         if (!cb->subbuf_start)
 484                 cb->subbuf_start = subbuf_start_default_callback;
 485         if (!cb->buf_mapped)
 486                 cb->buf_mapped = buf_mapped_default_callback;
 487         if (!cb->buf_unmapped)
 488                 cb->buf_unmapped = buf_unmapped_default_callback;
 489         if (!cb->create_buf_file)
 490                 cb->create_buf_file = create_buf_file_default_callback;
 491         if (!cb->remove_buf_file)
 492                 cb->remove_buf_file = remove_buf_file_default_callback;
 493         chan->cb = cb;
 494 }
 495
 496 /**
 497  *      relay_hotcpu_callback - CPU hotplug callback
 498  *      @nb: notifier block
 499  *      @action: hotplug action to take
 500  *      @hcpu: CPU number
 501  *
 502  *      Returns the success/failure of the operation. (%NOTIFY_OK, %NOTIFY_BAD)
 503  */
 504 static int __cpuinit relay_hotcpu_callback(struct notifier_block *nb,
 505                                 unsigned long action,
 506                                 void *hcpu)
 507 {
 508         unsigned int hotcpu = (unsigned long)hcpu;
 509         struct rchan *chan;
 510
 511         switch(action) {
 512         case CPU_UP_PREPARE:
 513         case CPU_UP_PREPARE_FROZEN:
 514                 mutex_lock(&relay_channels_mutex);
 515                 list_for_each_entry(chan, &relay_channels, list) {
 516                         if (chan->buf[hotcpu])
 517                                 continue;
 518                         chan->buf[hotcpu] = relay_open_buf(chan, hotcpu);
 519                         if(!chan->buf[hotcpu]) {
 520                                 printk(KERN_ERR
 521                                         "relay_hotcpu_callback: cpu %d buffer "
 522                                         "creation failed\n", hotcpu);
 523                                 mutex_unlock(&relay_channels_mutex);
 524                                 return NOTIFY_BAD;
 525                         }
 526                 }
 527                 mutex_unlock(&relay_channels_mutex);
 528                 break;
 529         case CPU_DEAD:
 530         case CPU_DEAD_FROZEN:
 531                 /* No need to flush the cpu : will be flushed upon
 532                  * final relay_flush() call. */
 533                 break;
 534         }
 535         return NOTIFY_OK;
 536 }
 537
 538 /**
 539  *      relay_open - create a new relay channel
 540  *      @base_filename: base name of files to create
 541  *      @parent: dentry of parent directory, %NULL for root directory
 542  *      @subbuf_size: size of sub-buffers
 543  *      @n_subbufs: number of sub-buffers
 544  *      @cb: client callback functions
 545  *      @private_data: user-defined data
 546  *
 547  *      Returns channel pointer if successful, %NULL otherwise.
 548  *
 549  *      Creates a channel buffer for each cpu using the sizes and
 550  *      attributes specified.  The created channel buffer files
 551  *      will be named base_filename0...base_filenameN-1.  File
 552  *      permissions will be %S_IRUSR.
 553  */
 554 struct rchan *relay_open(const char *base_filename,
 555                          struct dentry *parent,
 556                          size_t subbuf_size,
 557                          size_t n_subbufs,
 558                          struct rchan_callbacks *cb,
 559                          void *private_data)
 560 {
 561         unsigned int i;
 562         struct rchan *chan;
 563         if (!base_filename)
 564                 return NULL;
 565
 566         if (!(subbuf_size && n_subbufs))
 567                 return NULL;
 568
 569         chan = kzalloc(sizeof(struct rchan), GFP_KERNEL);
 570         if (!chan)
 571                 return NULL;
 572
 573         chan->version = RELAYFS_CHANNEL_VERSION;
 574         chan->n_subbufs = n_subbufs;
 575         chan->subbuf_size = subbuf_size;
 576         chan->alloc_size = FIX_SIZE(subbuf_size * n_subbufs);
 577         chan->parent = parent;
 578         chan->private_data = private_data;
 579         strlcpy(chan->base_filename, base_filename, NAME_MAX);
 580         setup_callbacks(chan, cb);
 581         kref_init(&chan->kref);
 582
 583         mutex_lock(&relay_channels_mutex);
 584         for_each_online_cpu(i) {
 585                 chan->buf[i] = relay_open_buf(chan, i);
 586                 if (!chan->buf[i])
 587                         goto free_bufs;
 588         }
 589         list_add(&chan->list, &relay_channels);
 590         mutex_unlock(&relay_channels_mutex);
 591
 592         return chan;
 593
 594 free_bufs:
 595         for_each_online_cpu(i) {
 596                 if (!chan->buf[i])
 597                         break;
 598                 relay_close_buf(chan->buf[i]);
 599         }
 600
 601         kref_put(&chan->kref, relay_destroy_channel);
 602         mutex_unlock(&relay_channels_mutex);
 603         return NULL;
 604 }
 605 EXPORT_SYMBOL_GPL(relay_open);
 606
 607 /**
 608  *      relay_switch_subbuf - switch to a new sub-buffer
 609  *      @buf: channel buffer
 610  *      @length: size of current event
 611  *
 612  *      Returns either the length passed in or 0 if full.
 613  *
 614  *      Performs sub-buffer-switch tasks such as invoking callbacks,
 615  *      updating padding counts, waking up readers, etc.
 616  */
 617 size_t relay_switch_subbuf(struct rchan_buf *buf, size_t length)
 618 {
 619         void *old, *new;
 620         size_t old_subbuf, new_subbuf;
 621
 622         if (unlikely(length > buf->chan->subbuf_size))
 623                 goto toobig;
 624
 625         if (buf->offset != buf->chan->subbuf_size + 1) {
 626                 buf->prev_padding = buf->chan->subbuf_size - buf->offset;
 627                 old_subbuf = buf->subbufs_produced % buf->chan->n_subbufs;
 628                 buf->padding[old_subbuf] = buf->prev_padding;
 629                 buf->subbufs_produced++;
 630                 buf->dentry->d_inode->i_size += buf->chan->subbuf_size -
 631                         buf->padding[old_subbuf];
 632                 smp_mb();
 633                 if (waitqueue_active(&buf->read_wait))
 634                         /*
 635                          * Calling wake_up_interruptible() from here
 636                          * will deadlock if we happen to be logging
 637                          * from the scheduler (trying to re-grab
 638                          * rq->lock), so defer it.
 639                          */
 640                         __mod_timer(&buf->timer, jiffies + 1);
 641         }
 642
 643         old = buf->data;
 644         new_subbuf = buf->subbufs_produced % buf->chan->n_subbufs;
 645         new = buf->start + new_subbuf * buf->chan->subbuf_size;
 646         buf->offset = 0;
 647         if (!buf->chan->cb->subbuf_start(buf, new, old, buf->prev_padding)) {
 648                 buf->offset = buf->chan->subbuf_size + 1;
 649                 return 0;
 650         }
 651         buf->data = new;
 652         buf->padding[new_subbuf] = 0;
 653
 654         if (unlikely(length + buf->offset > buf->chan->subbuf_size))
 655                 goto toobig;
 656
 657         return length;
 658
 659 toobig:
 660         buf->chan->last_toobig = length;
 661         return 0;
 662 }
 663 EXPORT_SYMBOL_GPL(relay_switch_subbuf);
 664
 665 /**
 666  *      relay_subbufs_consumed - update the buffer's sub-buffers-consumed count
 667  *      @chan: the channel
 668  *      @cpu: the cpu associated with the channel buffer to update
 669  *      @subbufs_consumed: number of sub-buffers to add to current buf's count
 670  *
 671  *      Adds to the channel buffer's consumed sub-buffer count.
 672  *      subbufs_consumed should be the number of sub-buffers newly consumed,
 673  *      not the total consumed.
 674  *
 675  *      NOTE. Kernel clients don't need to call this function if the channel
 676  *      mode is 'overwrite'.
 677  */
 678 void relay_subbufs_consumed(struct rchan *chan,
 679                             unsigned int cpu,
 680                             size_t subbufs_consumed)
 681 {
 682         struct rchan_buf *buf;
 683
 684         if (!chan)
 685                 return;
 686
 687         if (cpu >= NR_CPUS || !chan->buf[cpu])
 688                 return;
 689
 690         buf = chan->buf[cpu];
 691         buf->subbufs_consumed += subbufs_consumed;
 692         if (buf->subbufs_consumed > buf->subbufs_produced)
 693                 buf->subbufs_consumed = buf->subbufs_produced;
 694 }
 695 EXPORT_SYMBOL_GPL(relay_subbufs_consumed);
 696
 697 /**
 698  *      relay_close - close the channel
 699  *      @chan: the channel
 700  *
 701  *      Closes all channel buffers and frees the channel.
 702  */
 703 void relay_close(struct rchan *chan)
 704 {
 705         unsigned int i;
 706
 707         if (!chan)
 708                 return;
 709
 710         mutex_lock(&relay_channels_mutex);
 711         if (chan->is_global && chan->buf[0])
 712                 relay_close_buf(chan->buf[0]);
 713         else
 714                 for_each_possible_cpu(i)
 715                         if (chan->buf[i])
 716                                 relay_close_buf(chan->buf[i]);
 717
 718         if (chan->last_toobig)
 719                 printk(KERN_WARNING "relay: one or more items not logged "
 720                        "[item size (%Zd) > sub-buffer size (%Zd)]\n",
 721                        chan->last_toobig, chan->subbuf_size);
 722
 723         list_del(&chan->list);
 724         kref_put(&chan->kref, relay_destroy_channel);
 725         mutex_unlock(&relay_channels_mutex);
 726 }
 727 EXPORT_SYMBOL_GPL(relay_close);
 728
 729 /**
 730  *      relay_flush - close the channel
 731  *      @chan: the channel
 732  *
 733  *      Flushes all channel buffers, i.e. forces buffer switch.
 734  */
 735 void relay_flush(struct rchan *chan)
 736 {
 737         unsigned int i;
 738
 739         if (!chan)
 740                 return;
 741
 742         if (chan->is_global && chan->buf[0]) {
 743                 relay_switch_subbuf(chan->buf[0], 0);
 744                 return;
 745         }
 746
 747         mutex_lock(&relay_channels_mutex);
 748         for_each_possible_cpu(i)
 749                 if (chan->buf[i])
 750                         relay_switch_subbuf(chan->buf[i], 0);
 751         mutex_unlock(&relay_channels_mutex);
 752 }
 753 EXPORT_SYMBOL_GPL(relay_flush);
 754
 755 /**
 756  *      relay_file_open - open file op for relay files
 757  *      @inode: the inode
 758  *      @filp: the file
 759  *
 760  *      Increments the channel buffer refcount.
 761  */
 762 static int relay_file_open(struct inode *inode, struct file *filp)
 763 {
 764         struct rchan_buf *buf = inode->i_private;
 765         kref_get(&buf->kref);
 766         filp->private_data = buf;
 767
 768         return nonseekable_open(inode, filp);
 769 }
 770
 771 /**
 772  *      relay_file_mmap - mmap file op for relay files
 773  *      @filp: the file
 774  *      @vma: the vma describing what to map
 775  *
 776  *      Calls upon relay_mmap_buf() to map the file into user space.
 777  */
 778 static int relay_file_mmap(struct file *filp, struct vm_area_struct *vma)
 779 {
 780         struct rchan_buf *buf = filp->private_data;
 781         return relay_mmap_buf(buf, vma);
 782 }
 783
 784 /**
 785  *      relay_file_poll - poll file op for relay files
 786  *      @filp: the file
 787  *      @wait: poll table
 788  *
 789  *      Poll implemention.
 790  */
 791 static unsigned int relay_file_poll(struct file *filp, poll_table *wait)
 792 {
 793         unsigned int mask = 0;
 794         struct rchan_buf *buf = filp->private_data;
 795
 796         if (buf->finalized)
 797                 return POLLERR;
 798
 799         if (filp->f_mode & FMODE_READ) {
 800                 poll_wait(filp, &buf->read_wait, wait);
 801                 if (!relay_buf_empty(buf))
 802                         mask |= POLLIN | POLLRDNORM;
 803         }
 804
 805         return mask;
 806 }
 807
 808 /**
 809  *      relay_file_release - release file op for relay files
 810  *      @inode: the inode
 811  *      @filp: the file
 812  *
 813  *      Decrements the channel refcount, as the filesystem is
 814  *      no longer using it.
 815  */
 816 static int relay_file_release(struct inode *inode, struct file *filp)
 817 {
 818         struct rchan_buf *buf = filp->private_data;
 819         kref_put(&buf->kref, relay_remove_buf);
 820
 821         return 0;
 822 }
 823
 824 /*
 825  *      relay_file_read_consume - update the consumed count for the buffer
 826  */
 827 static void relay_file_read_consume(struct rchan_buf *buf,
 828                                     size_t read_pos,
 829                                     size_t bytes_consumed)
 830 {
 831         size_t subbuf_size = buf->chan->subbuf_size;
 832         size_t n_subbufs = buf->chan->n_subbufs;
 833         size_t read_subbuf;
 834
 835         if (buf->bytes_consumed + bytes_consumed > subbuf_size) {
 836                 relay_subbufs_consumed(buf->chan, buf->cpu, 1);
 837                 buf->bytes_consumed = 0;
 838         }
 839
 840         buf->bytes_consumed += bytes_consumed;
 841         if (!read_pos)
 842                 read_subbuf = buf->subbufs_consumed % n_subbufs;
 843         else
 844                 read_subbuf = read_pos / buf->chan->subbuf_size;
 845         if (buf->bytes_consumed + buf->padding[read_subbuf] == subbuf_size) {
 846                 if ((read_subbuf == buf->subbufs_produced % n_subbufs) &&
 847                     (buf->offset == subbuf_size))
 848                         return;
 849                 relay_subbufs_consumed(buf->chan, buf->cpu, 1);
 850                 buf->bytes_consumed = 0;
 851         }
 852 }
 853
 854 /*
 855  *      relay_file_read_avail - boolean, are there unconsumed bytes available?
 856  */
 857 static int relay_file_read_avail(struct rchan_buf *buf, size_t read_pos)
 858 {
 859         size_t subbuf_size = buf->chan->subbuf_size;
 860         size_t n_subbufs = buf->chan->n_subbufs;
 861         size_t produced = buf->subbufs_produced;
 862         size_t consumed = buf->subbufs_consumed;
 863
 864         relay_file_read_consume(buf, read_pos, 0);
 865
 866         if (unlikely(buf->offset > subbuf_size)) {
 867                 if (produced == consumed)
 868                         return 0;
 869                 return 1;
 870         }
 871
 872         if (unlikely(produced - consumed >= n_subbufs)) {
 873                 consumed = produced - n_subbufs + 1;
 874                 buf->subbufs_consumed = consumed;
 875                 buf->bytes_consumed = 0;
 876         }
 877
 878         produced = (produced % n_subbufs) * subbuf_size + buf->offset;
 879         consumed = (consumed % n_subbufs) * subbuf_size + buf->bytes_consumed;
 880
 881         if (consumed > produced)
 882                 produced += n_subbufs * subbuf_size;
 883
 884         if (consumed == produced)
 885                 return 0;
 886
 887         return 1;
 888 }
 889
 890 /**
 891  *      relay_file_read_subbuf_avail - return bytes available in sub-buffer
 892  *      @read_pos: file read position
 893  *      @buf: relay channel buffer
 894  */
 895 static size_t relay_file_read_subbuf_avail(size_t read_pos,
 896                                            struct rchan_buf *buf)
 897 {
 898         size_t padding, avail = 0;
 899         size_t read_subbuf, read_offset, write_subbuf, write_offset;
 900         size_t subbuf_size = buf->chan->subbuf_size;
 901
 902         write_subbuf = (buf->data - buf->start) / subbuf_size;
 903         write_offset = buf->offset > subbuf_size ? subbuf_size : buf->offset;
 904         read_subbuf = read_pos / subbuf_size;
 905         read_offset = read_pos % subbuf_size;
 906         padding = buf->padding[read_subbuf];
 907
 908         if (read_subbuf == write_subbuf) {
 909                 if (read_offset + padding < write_offset)
 910                         avail = write_offset - (read_offset + padding);
 911         } else
 912                 avail = (subbuf_size - padding) - read_offset;
 913
 914         return avail;
 915 }
 916
 917 /**
 918  *      relay_file_read_start_pos - find the first available byte to read
 919  *      @read_pos: file read position
 920  *      @buf: relay channel buffer
 921  *
 922  *      If the @read_pos is in the middle of padding, return the
 923  *      position of the first actually available byte, otherwise
 924  *      return the original value.
 925  */
 926 static size_t relay_file_read_start_pos(size_t read_pos,
 927                                         struct rchan_buf *buf)
 928 {
 929         size_t read_subbuf, padding, padding_start, padding_end;
 930         size_t subbuf_size = buf->chan->subbuf_size;
 931         size_t n_subbufs = buf->chan->n_subbufs;
 932         size_t consumed = buf->subbufs_consumed % n_subbufs;
 933
 934         if (!read_pos)
 935                 read_pos = consumed * subbuf_size + buf->bytes_consumed;
 936         read_subbuf = read_pos / subbuf_size;
 937         padding = buf->padding[read_subbuf];
 938         padding_start = (read_subbuf + 1) * subbuf_size - padding;
 939         padding_end = (read_subbuf + 1) * subbuf_size;
 940         if (read_pos >= padding_start && read_pos < padding_end) {
 941                 read_subbuf = (read_subbuf + 1) % n_subbufs;
 942                 read_pos = read_subbuf * subbuf_size;
 943         }
 944
 945         return read_pos;
 946 }
 947
 948 /**
 949  *      relay_file_read_end_pos - return the new read position
 950  *      @read_pos: file read position
 951  *      @buf: relay channel buffer
 952  *      @count: number of bytes to be read
 953  */
 954 static size_t relay_file_read_end_pos(struct rchan_buf *buf,
 955                                       size_t read_pos,
 956                                       size_t count)
 957 {
 958         size_t read_subbuf, padding, end_pos;
 959         size_t subbuf_size = buf->chan->subbuf_size;
 960         size_t n_subbufs = buf->chan->n_subbufs;
 961
 962         read_subbuf = read_pos / subbuf_size;
 963         padding = buf->padding[read_subbuf];
 964         if (read_pos % subbuf_size + count + padding == subbuf_size)
 965                 end_pos = (read_subbuf + 1) * subbuf_size;
 966         else
 967                 end_pos = read_pos + count;
 968         if (end_pos >= subbuf_size * n_subbufs)
 969                 end_pos = 0;
 970
 971         return end_pos;
 972 }
 973
 974 /*
 975  *      subbuf_read_actor - read up to one subbuf's worth of data
 976  */
 977 static int subbuf_read_actor(size_t read_start,
 978                              struct rchan_buf *buf,
 979                              size_t avail,
 980                              read_descriptor_t *desc,
 981                              read_actor_t actor)
 982 {
 983         void *from;
 984         int ret = 0;
 985
 986         from = buf->start + read_start;
 987         ret = avail;
 988         if (copy_to_user(desc->arg.buf, from, avail)) {
 989                 desc->error = -EFAULT;
 990                 ret = 0;
 991         }
 992         desc->arg.data += ret;
 993         desc->written += ret;
 994         desc->count -= ret;
 995
 996         return ret;
 997 }
 998
 999 typedef int (*subbuf_actor_t) (size_t read_start,
1000                                struct rchan_buf *buf,
1001                                size_t avail,
1002                                read_descriptor_t *desc,
1003                                read_actor_t actor);
1004
1005 /*
1006  *      relay_file_read_subbufs - read count bytes, bridging subbuf boundaries
1007  */
1008 static ssize_t relay_file_read_subbufs(struct file *filp, loff_t *ppos,
1009                                         subbuf_actor_t subbuf_actor,
1010                                         read_actor_t actor,
1011                                         read_descriptor_t *desc)
1012 {
1013         struct rchan_buf *buf = filp->private_data;
1014         size_t read_start, avail;
1015         int ret;
1016
1017         if (!desc->count)
1018                 return 0;
1019
1020         mutex_lock(&filp->f_path.dentry->d_inode->i_mutex);
1021         do {
1022                 if (!relay_file_read_avail(buf, *ppos))
1023                         break;
1024
1025                 read_start = relay_file_read_start_pos(*ppos, buf);
1026                 avail = relay_file_read_subbuf_avail(read_start, buf);
1027                 if (!avail)
1028                         break;
1029
1030                 avail = min(desc->count, avail);
1031                 ret = subbuf_actor(read_start, buf, avail, desc, actor);
1032                 if (desc->error < 0)
1033                         break;
1034
1035                 if (ret) {
1036                         relay_file_read_consume(buf, read_start, ret);
1037                         *ppos = relay_file_read_end_pos(buf, read_start, ret);
1038                 }
1039         } while (desc->count && ret);
1040         mutex_unlock(&filp->f_path.dentry->d_inode->i_mutex);
1041
1042         return desc->written;
1043 }
1044
1045 static ssize_t relay_file_read(struct file *filp,
1046                                char __user *buffer,
1047                                size_t count,
1048                                loff_t *ppos)
1049 {
1050         read_descriptor_t desc;
1051         desc.written = 0;
1052         desc.count = count;
1053         desc.arg.buf = buffer;
1054         desc.error = 0;
1055         return relay_file_read_subbufs(filp, ppos, subbuf_read_actor,
1056                                        NULL, &desc);
1057 }
1058
1059 static void relay_consume_bytes(struct rchan_buf *rbuf, int bytes_consumed)
1060 {
1061         rbuf->bytes_consumed += bytes_consumed;
1062
1063         if (rbuf->bytes_consumed >= rbuf->chan->subbuf_size) {
1064                 relay_subbufs_consumed(rbuf->chan, rbuf->cpu, 1);
1065                 rbuf->bytes_consumed %= rbuf->chan->subbuf_size;
1066         }
1067 }
1068
1069 static void relay_pipe_buf_release(struct pipe_inode_info *pipe,
1070                                    struct pipe_buffer *buf)
1071 {
1072         struct rchan_buf *rbuf;
1073
1074         rbuf = (struct rchan_buf *)page_private(buf->page);
1075         relay_consume_bytes(rbuf, buf->private);
1076 }
1077
1078 static struct pipe_buf_operations relay_pipe_buf_ops = {
1079         .can_merge = 0,
1080         .map = generic_pipe_buf_map,
1081         .unmap = generic_pipe_buf_unmap,
1082         .confirm = generic_pipe_buf_confirm,
1083         .release = relay_pipe_buf_release,
1084         .steal = generic_pipe_buf_steal,
1085         .get = generic_pipe_buf_get,
1086 };
1087
1088 static void relay_page_release(struct splice_pipe_desc *spd, unsigned int i)
1089 {
1090 }
1091
1092 /*
1093  *      subbuf_splice_actor - splice up to one subbuf's worth of data
1094  */
1095 static int subbuf_splice_actor(struct file *in,
1096                                loff_t *ppos,
1097                                struct pipe_inode_info *pipe,
1098                                size_t len,
1099                                unsigned int flags,
1100                                int *nonpad_ret)
1101 {
1102         unsigned int pidx, poff, total_len, subbuf_pages, nr_pages, ret;
1103         struct rchan_buf *rbuf = in->private_data;
1104         unsigned int subbuf_size = rbuf->chan->subbuf_size;
1105         uint64_t pos = (uint64_t) *ppos;
1106         uint32_t alloc_size = (uint32_t) rbuf->chan->alloc_size;
1107         size_t read_start = (size_t) do_div(pos, alloc_size);
1108         size_t read_subbuf = read_start / subbuf_size;
1109         size_t padding = rbuf->padding[read_subbuf];
1110         size_t nonpad_end = read_subbuf * subbuf_size + subbuf_size - padding;
1111         struct page *pages[PIPE_BUFFERS];
1112         struct partial_page partial[PIPE_BUFFERS];
1113         struct splice_pipe_desc spd = {
1114                 .pages = pages,
1115                 .nr_pages = 0,
1116                 .partial = partial,
1117                 .flags = flags,
1118                 .ops = &relay_pipe_buf_ops,
1119                 .spd_release = relay_page_release,
1120         };
1121
1122         if (rbuf->subbufs_produced == rbuf->subbufs_consumed)
1123                 return 0;
1124
1125         /*
1126          * Adjust read len, if longer than what is available
1127          */
1128         if (len > (subbuf_size - read_start % subbuf_size))
1129                 len = subbuf_size - read_start % subbuf_size;
1130
1131         subbuf_pages = rbuf->chan->alloc_size >> PAGE_SHIFT;
1132         pidx = (read_start / PAGE_SIZE) % subbuf_pages;
1133         poff = read_start & ~PAGE_MASK;
1134         nr_pages = min_t(unsigned int, subbuf_pages, PIPE_BUFFERS);
1135
1136         for (total_len = 0; spd.nr_pages < nr_pages; spd.nr_pages++) {
1137                 unsigned int this_len, this_end, private;
1138                 unsigned int cur_pos = read_start + total_len;
1139
1140                 if (!len)
1141                         break;
1142
1143                 this_len = min_t(unsigned long, len, PAGE_SIZE - poff);
1144                 private = this_len;
1145
1146                 spd.pages[spd.nr_pages] = rbuf->page_array[pidx];
1147                 spd.partial[spd.nr_pages].offset = poff;
1148
1149                 this_end = cur_pos + this_len;
1150                 if (this_end >= nonpad_end) {
1151                         this_len = nonpad_end - cur_pos;
1152                         private = this_len + padding;
1153                 }
1154                 spd.partial[spd.nr_pages].len = this_len;
1155                 spd.partial[spd.nr_pages].private = private;
1156
1157                 len -= this_len;
1158                 total_len += this_len;
1159                 poff = 0;
1160                 pidx = (pidx + 1) % subbuf_pages;
1161
1162                 if (this_end >= nonpad_end) {
1163                         spd.nr_pages++;
1164                         break;
1165                 }
1166         }
1167
1168         if (!spd.nr_pages)
1169                 return 0;
1170
1171         ret = *nonpad_ret = splice_to_pipe(pipe, &spd);
1172         if (ret < 0 || ret < total_len)
1173                 return ret;
1174
1175         if (read_start + ret == nonpad_end)
1176                 ret += padding;
1177
1178         return ret;
1179 }
1180
1181 static ssize_t relay_file_splice_read(struct file *in,
1182                                       loff_t *ppos,
1183                                       struct pipe_inode_info *pipe,
1184                                       size_t len,
1185                                       unsigned int flags)
1186 {
1187         ssize_t spliced;
1188         int ret;
1189         int nonpad_ret = 0;
1190
1191         ret = 0;
1192         spliced = 0;
1193
1194         while (len && !spliced) {
1195                 ret = subbuf_splice_actor(in, ppos, pipe, len, flags, &nonpad_ret);
1196                 if (ret < 0)
1197                         break;
1198                 else if (!ret) {
1199                         if (spliced)
1200                                 break;
1201                         if (flags & SPLICE_F_NONBLOCK) {
1202                                 ret = -EAGAIN;
1203                                 break;
1204                         }
1205                 }
1206
1207                 *ppos += ret;
1208                 if (ret > len)
1209                         len = 0;
1210                 else
1211                         len -= ret;
1212                 spliced += nonpad_ret;
1213                 nonpad_ret = 0;
1214         }
1215
1216         if (spliced)
1217                 return spliced;
1218
1219         return ret;
1220 }
1221
1222 const struct file_operations relay_file_operations = {
1223         .open           = relay_file_open,
1224         .poll           = relay_file_poll,
1225         .mmap           = relay_file_mmap,
1226         .read           = relay_file_read,
1227         .llseek         = no_llseek,
1228         .release        = relay_file_release,
1229         .splice_read    = relay_file_splice_read,
1230 };
1231 EXPORT_SYMBOL_GPL(relay_file_operations);
1232
1233 static __init int relay_init(void)
1234 {
1235
1236         hotcpu_notifier(relay_hotcpu_callback, 0);
1237         return 0;
1238 }
1239
1240 module_init(relay_init);