linux/kernel/relay.c
<<
>>
Prefs
   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 */
  27static DEFINE_MUTEX(relay_channels_mutex);
  28static LIST_HEAD(relay_channels);
  29
  30/*
  31 * close() vm_op implementation for relay file mapping.
  32 */
  33static 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 */
  42static 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 */
  63static const 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 */
  71static 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 */
  89static 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 */
 106static 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 */
 133static 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
 159depopulate:
 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 */
 172static 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
 190free_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 */
 202static 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 */
 212static 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 */
 237static 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 */
 250static 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 */
 261int 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}
 266EXPORT_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 */
 280static 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 */
 294static 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 */
 302static 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 */
 310static 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 */
 322static int remove_buf_file_default_callback(struct dentry *dentry)
 323{
 324        return -EINVAL;
 325}
 326
 327/* relay channel default callbacks */
 328static 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 */
 342static 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 */
 355static 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 */
 390void 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_possible_cpu(i)
 404                if (chan->buf[i])
 405                        __relay_reset(chan->buf[i], 0);
 406        mutex_unlock(&relay_channels_mutex);
 407}
 408EXPORT_SYMBOL_GPL(relay_reset);
 409
 410static inline void relay_set_buf_dentry(struct rchan_buf *buf,
 411                                        struct dentry *dentry)
 412{
 413        buf->dentry = dentry;
 414        buf->dentry->d_inode->i_size = buf->early_bytes;
 415}
 416
 417static struct dentry *relay_create_buf_file(struct rchan *chan,
 418                                            struct rchan_buf *buf,
 419                                            unsigned int cpu)
 420{
 421        struct dentry *dentry;
 422        char *tmpname;
 423
 424        tmpname = kzalloc(NAME_MAX + 1, GFP_KERNEL);
 425        if (!tmpname)
 426                return NULL;
 427        snprintf(tmpname, NAME_MAX, "%s%d", chan->base_filename, cpu);
 428
 429        /* Create file in fs */
 430        dentry = chan->cb->create_buf_file(tmpname, chan->parent,
 431                                           S_IRUSR, buf,
 432                                           &chan->is_global);
 433
 434        kfree(tmpname);
 435
 436        return dentry;
 437}
 438
 439/*
 440 *      relay_open_buf - create a new relay channel buffer
 441 *
 442 *      used by relay_open() and CPU hotplug.
 443 */
 444static struct rchan_buf *relay_open_buf(struct rchan *chan, unsigned int cpu)
 445{
 446        struct rchan_buf *buf = NULL;
 447        struct dentry *dentry;
 448
 449        if (chan->is_global)
 450                return chan->buf[0];
 451
 452        buf = relay_create_buf(chan);
 453        if (!buf)
 454                return NULL;
 455
 456        if (chan->has_base_filename) {
 457                dentry = relay_create_buf_file(chan, buf, cpu);
 458                if (!dentry)
 459                        goto free_buf;
 460                relay_set_buf_dentry(buf, dentry);
 461        }
 462
 463        buf->cpu = cpu;
 464        __relay_reset(buf, 1);
 465
 466        if(chan->is_global) {
 467                chan->buf[0] = buf;
 468                buf->cpu = 0;
 469        }
 470
 471        return buf;
 472
 473free_buf:
 474        relay_destroy_buf(buf);
 475        return NULL;
 476}
 477
 478/**
 479 *      relay_close_buf - close a channel buffer
 480 *      @buf: channel buffer
 481 *
 482 *      Marks the buffer finalized and restores the default callbacks.
 483 *      The channel buffer and channel buffer data structure are then freed
 484 *      automatically when the last reference is given up.
 485 */
 486static void relay_close_buf(struct rchan_buf *buf)
 487{
 488        buf->finalized = 1;
 489        del_timer_sync(&buf->timer);
 490        kref_put(&buf->kref, relay_remove_buf);
 491}
 492
 493static void setup_callbacks(struct rchan *chan,
 494                                   struct rchan_callbacks *cb)
 495{
 496        if (!cb) {
 497                chan->cb = &default_channel_callbacks;
 498                return;
 499        }
 500
 501        if (!cb->subbuf_start)
 502                cb->subbuf_start = subbuf_start_default_callback;
 503        if (!cb->buf_mapped)
 504                cb->buf_mapped = buf_mapped_default_callback;
 505        if (!cb->buf_unmapped)
 506                cb->buf_unmapped = buf_unmapped_default_callback;
 507        if (!cb->create_buf_file)
 508                cb->create_buf_file = create_buf_file_default_callback;
 509        if (!cb->remove_buf_file)
 510                cb->remove_buf_file = remove_buf_file_default_callback;
 511        chan->cb = cb;
 512}
 513
 514/**
 515 *      relay_hotcpu_callback - CPU hotplug callback
 516 *      @nb: notifier block
 517 *      @action: hotplug action to take
 518 *      @hcpu: CPU number
 519 *
 520 *      Returns the success/failure of the operation. (%NOTIFY_OK, %NOTIFY_BAD)
 521 */
 522static int __cpuinit relay_hotcpu_callback(struct notifier_block *nb,
 523                                unsigned long action,
 524                                void *hcpu)
 525{
 526        unsigned int hotcpu = (unsigned long)hcpu;
 527        struct rchan *chan;
 528
 529        switch(action) {
 530        case CPU_UP_PREPARE:
 531        case CPU_UP_PREPARE_FROZEN:
 532                mutex_lock(&relay_channels_mutex);
 533                list_for_each_entry(chan, &relay_channels, list) {
 534                        if (chan->buf[hotcpu])
 535                                continue;
 536                        chan->buf[hotcpu] = relay_open_buf(chan, hotcpu);
 537                        if(!chan->buf[hotcpu]) {
 538                                printk(KERN_ERR
 539                                        "relay_hotcpu_callback: cpu %d buffer "
 540                                        "creation failed\n", hotcpu);
 541                                mutex_unlock(&relay_channels_mutex);
 542                                return NOTIFY_BAD;
 543                        }
 544                }
 545                mutex_unlock(&relay_channels_mutex);
 546                break;
 547        case CPU_DEAD:
 548        case CPU_DEAD_FROZEN:
 549                /* No need to flush the cpu : will be flushed upon
 550                 * final relay_flush() call. */
 551                break;
 552        }
 553        return NOTIFY_OK;
 554}
 555
 556/**
 557 *      relay_open - create a new relay channel
 558 *      @base_filename: base name of files to create, %NULL for buffering only
 559 *      @parent: dentry of parent directory, %NULL for root directory or buffer
 560 *      @subbuf_size: size of sub-buffers
 561 *      @n_subbufs: number of sub-buffers
 562 *      @cb: client callback functions
 563 *      @private_data: user-defined data
 564 *
 565 *      Returns channel pointer if successful, %NULL otherwise.
 566 *
 567 *      Creates a channel buffer for each cpu using the sizes and
 568 *      attributes specified.  The created channel buffer files
 569 *      will be named base_filename0...base_filenameN-1.  File
 570 *      permissions will be %S_IRUSR.
 571 */
 572struct rchan *relay_open(const char *base_filename,
 573                         struct dentry *parent,
 574                         size_t subbuf_size,
 575                         size_t n_subbufs,
 576                         struct rchan_callbacks *cb,
 577                         void *private_data)
 578{
 579        unsigned int i;
 580        struct rchan *chan;
 581
 582        if (!(subbuf_size && n_subbufs))
 583                return NULL;
 584
 585        chan = kzalloc(sizeof(struct rchan), GFP_KERNEL);
 586        if (!chan)
 587                return NULL;
 588
 589        chan->version = RELAYFS_CHANNEL_VERSION;
 590        chan->n_subbufs = n_subbufs;
 591        chan->subbuf_size = subbuf_size;
 592        chan->alloc_size = FIX_SIZE(subbuf_size * n_subbufs);
 593        chan->parent = parent;
 594        chan->private_data = private_data;
 595        if (base_filename) {
 596                chan->has_base_filename = 1;
 597                strlcpy(chan->base_filename, base_filename, NAME_MAX);
 598        }
 599        setup_callbacks(chan, cb);
 600        kref_init(&chan->kref);
 601
 602        mutex_lock(&relay_channels_mutex);
 603        for_each_online_cpu(i) {
 604                chan->buf[i] = relay_open_buf(chan, i);
 605                if (!chan->buf[i])
 606                        goto free_bufs;
 607        }
 608        list_add(&chan->list, &relay_channels);
 609        mutex_unlock(&relay_channels_mutex);
 610
 611        return chan;
 612
 613free_bufs:
 614        for_each_possible_cpu(i) {
 615                if (chan->buf[i])
 616                        relay_close_buf(chan->buf[i]);
 617        }
 618
 619        kref_put(&chan->kref, relay_destroy_channel);
 620        mutex_unlock(&relay_channels_mutex);
 621        return NULL;
 622}
 623EXPORT_SYMBOL_GPL(relay_open);
 624
 625struct rchan_percpu_buf_dispatcher {
 626        struct rchan_buf *buf;
 627        struct dentry *dentry;
 628};
 629
 630/* Called in atomic context. */
 631static void __relay_set_buf_dentry(void *info)
 632{
 633        struct rchan_percpu_buf_dispatcher *p = info;
 634
 635        relay_set_buf_dentry(p->buf, p->dentry);
 636}
 637
 638/**
 639 *      relay_late_setup_files - triggers file creation
 640 *      @chan: channel to operate on
 641 *      @base_filename: base name of files to create
 642 *      @parent: dentry of parent directory, %NULL for root directory
 643 *
 644 *      Returns 0 if successful, non-zero otherwise.
 645 *
 646 *      Use to setup files for a previously buffer-only channel.
 647 *      Useful to do early tracing in kernel, before VFS is up, for example.
 648 */
 649int relay_late_setup_files(struct rchan *chan,
 650                           const char *base_filename,
 651                           struct dentry *parent)
 652{
 653        int err = 0;
 654        unsigned int i, curr_cpu;
 655        unsigned long flags;
 656        struct dentry *dentry;
 657        struct rchan_percpu_buf_dispatcher disp;
 658
 659        if (!chan || !base_filename)
 660                return -EINVAL;
 661
 662        strlcpy(chan->base_filename, base_filename, NAME_MAX);
 663
 664        mutex_lock(&relay_channels_mutex);
 665        /* Is chan already set up? */
 666        if (unlikely(chan->has_base_filename)) {
 667                mutex_unlock(&relay_channels_mutex);
 668                return -EEXIST;
 669        }
 670        chan->has_base_filename = 1;
 671        chan->parent = parent;
 672        curr_cpu = get_cpu();
 673        /*
 674         * The CPU hotplug notifier ran before us and created buffers with
 675         * no files associated. So it's safe to call relay_setup_buf_file()
 676         * on all currently online CPUs.
 677         */
 678        for_each_online_cpu(i) {
 679                if (unlikely(!chan->buf[i])) {
 680                        WARN_ONCE(1, KERN_ERR "CPU has no buffer!\n");
 681                        err = -EINVAL;
 682                        break;
 683                }
 684
 685                dentry = relay_create_buf_file(chan, chan->buf[i], i);
 686                if (unlikely(!dentry)) {
 687                        err = -EINVAL;
 688                        break;
 689                }
 690
 691                if (curr_cpu == i) {
 692                        local_irq_save(flags);
 693                        relay_set_buf_dentry(chan->buf[i], dentry);
 694                        local_irq_restore(flags);
 695                } else {
 696                        disp.buf = chan->buf[i];
 697                        disp.dentry = dentry;
 698                        smp_mb();
 699                        /* relay_channels_mutex must be held, so wait. */
 700                        err = smp_call_function_single(i,
 701                                                       __relay_set_buf_dentry,
 702                                                       &disp, 1);
 703                }
 704                if (unlikely(err))
 705                        break;
 706        }
 707        put_cpu();
 708        mutex_unlock(&relay_channels_mutex);
 709
 710        return err;
 711}
 712
 713/**
 714 *      relay_switch_subbuf - switch to a new sub-buffer
 715 *      @buf: channel buffer
 716 *      @length: size of current event
 717 *
 718 *      Returns either the length passed in or 0 if full.
 719 *
 720 *      Performs sub-buffer-switch tasks such as invoking callbacks,
 721 *      updating padding counts, waking up readers, etc.
 722 */
 723size_t relay_switch_subbuf(struct rchan_buf *buf, size_t length)
 724{
 725        void *old, *new;
 726        size_t old_subbuf, new_subbuf;
 727
 728        if (unlikely(length > buf->chan->subbuf_size))
 729                goto toobig;
 730
 731        if (buf->offset != buf->chan->subbuf_size + 1) {
 732                buf->prev_padding = buf->chan->subbuf_size - buf->offset;
 733                old_subbuf = buf->subbufs_produced % buf->chan->n_subbufs;
 734                buf->padding[old_subbuf] = buf->prev_padding;
 735                buf->subbufs_produced++;
 736                if (buf->dentry)
 737                        buf->dentry->d_inode->i_size +=
 738                                buf->chan->subbuf_size -
 739                                buf->padding[old_subbuf];
 740                else
 741                        buf->early_bytes += buf->chan->subbuf_size -
 742                                            buf->padding[old_subbuf];
 743                smp_mb();
 744                if (waitqueue_active(&buf->read_wait))
 745                        /*
 746                         * Calling wake_up_interruptible() from here
 747                         * will deadlock if we happen to be logging
 748                         * from the scheduler (trying to re-grab
 749                         * rq->lock), so defer it.
 750                         */
 751                        mod_timer(&buf->timer, jiffies + 1);
 752        }
 753
 754        old = buf->data;
 755        new_subbuf = buf->subbufs_produced % buf->chan->n_subbufs;
 756        new = buf->start + new_subbuf * buf->chan->subbuf_size;
 757        buf->offset = 0;
 758        if (!buf->chan->cb->subbuf_start(buf, new, old, buf->prev_padding)) {
 759                buf->offset = buf->chan->subbuf_size + 1;
 760                return 0;
 761        }
 762        buf->data = new;
 763        buf->padding[new_subbuf] = 0;
 764
 765        if (unlikely(length + buf->offset > buf->chan->subbuf_size))
 766                goto toobig;
 767
 768        return length;
 769
 770toobig:
 771        buf->chan->last_toobig = length;
 772        return 0;
 773}
 774EXPORT_SYMBOL_GPL(relay_switch_subbuf);
 775
 776/**
 777 *      relay_subbufs_consumed - update the buffer's sub-buffers-consumed count
 778 *      @chan: the channel
 779 *      @cpu: the cpu associated with the channel buffer to update
 780 *      @subbufs_consumed: number of sub-buffers to add to current buf's count
 781 *
 782 *      Adds to the channel buffer's consumed sub-buffer count.
 783 *      subbufs_consumed should be the number of sub-buffers newly consumed,
 784 *      not the total consumed.
 785 *
 786 *      NOTE. Kernel clients don't need to call this function if the channel
 787 *      mode is 'overwrite'.
 788 */
 789void relay_subbufs_consumed(struct rchan *chan,
 790                            unsigned int cpu,
 791                            size_t subbufs_consumed)
 792{
 793        struct rchan_buf *buf;
 794
 795        if (!chan)
 796                return;
 797
 798        if (cpu >= NR_CPUS || !chan->buf[cpu] ||
 799                                        subbufs_consumed > chan->n_subbufs)
 800                return;
 801
 802        buf = chan->buf[cpu];
 803        if (subbufs_consumed > buf->subbufs_produced - buf->subbufs_consumed)
 804                buf->subbufs_consumed = buf->subbufs_produced;
 805        else
 806                buf->subbufs_consumed += subbufs_consumed;
 807}
 808EXPORT_SYMBOL_GPL(relay_subbufs_consumed);
 809
 810/**
 811 *      relay_close - close the channel
 812 *      @chan: the channel
 813 *
 814 *      Closes all channel buffers and frees the channel.
 815 */
 816void relay_close(struct rchan *chan)
 817{
 818        unsigned int i;
 819
 820        if (!chan)
 821                return;
 822
 823        mutex_lock(&relay_channels_mutex);
 824        if (chan->is_global && chan->buf[0])
 825                relay_close_buf(chan->buf[0]);
 826        else
 827                for_each_possible_cpu(i)
 828                        if (chan->buf[i])
 829                                relay_close_buf(chan->buf[i]);
 830
 831        if (chan->last_toobig)
 832                printk(KERN_WARNING "relay: one or more items not logged "
 833                       "[item size (%Zd) > sub-buffer size (%Zd)]\n",
 834                       chan->last_toobig, chan->subbuf_size);
 835
 836        list_del(&chan->list);
 837        kref_put(&chan->kref, relay_destroy_channel);
 838        mutex_unlock(&relay_channels_mutex);
 839}
 840EXPORT_SYMBOL_GPL(relay_close);
 841
 842/**
 843 *      relay_flush - close the channel
 844 *      @chan: the channel
 845 *
 846 *      Flushes all channel buffers, i.e. forces buffer switch.
 847 */
 848void relay_flush(struct rchan *chan)
 849{
 850        unsigned int i;
 851
 852        if (!chan)
 853                return;
 854
 855        if (chan->is_global && chan->buf[0]) {
 856                relay_switch_subbuf(chan->buf[0], 0);
 857                return;
 858        }
 859
 860        mutex_lock(&relay_channels_mutex);
 861        for_each_possible_cpu(i)
 862                if (chan->buf[i])
 863                        relay_switch_subbuf(chan->buf[i], 0);
 864        mutex_unlock(&relay_channels_mutex);
 865}
 866EXPORT_SYMBOL_GPL(relay_flush);
 867
 868/**
 869 *      relay_file_open - open file op for relay files
 870 *      @inode: the inode
 871 *      @filp: the file
 872 *
 873 *      Increments the channel buffer refcount.
 874 */
 875static int relay_file_open(struct inode *inode, struct file *filp)
 876{
 877        struct rchan_buf *buf = inode->i_private;
 878        kref_get(&buf->kref);
 879        filp->private_data = buf;
 880
 881        return nonseekable_open(inode, filp);
 882}
 883
 884/**
 885 *      relay_file_mmap - mmap file op for relay files
 886 *      @filp: the file
 887 *      @vma: the vma describing what to map
 888 *
 889 *      Calls upon relay_mmap_buf() to map the file into user space.
 890 */
 891static int relay_file_mmap(struct file *filp, struct vm_area_struct *vma)
 892{
 893        struct rchan_buf *buf = filp->private_data;
 894        return relay_mmap_buf(buf, vma);
 895}
 896
 897/**
 898 *      relay_file_poll - poll file op for relay files
 899 *      @filp: the file
 900 *      @wait: poll table
 901 *
 902 *      Poll implemention.
 903 */
 904static unsigned int relay_file_poll(struct file *filp, poll_table *wait)
 905{
 906        unsigned int mask = 0;
 907        struct rchan_buf *buf = filp->private_data;
 908
 909        if (buf->finalized)
 910                return POLLERR;
 911
 912        if (filp->f_mode & FMODE_READ) {
 913                poll_wait(filp, &buf->read_wait, wait);
 914                if (!relay_buf_empty(buf))
 915                        mask |= POLLIN | POLLRDNORM;
 916        }
 917
 918        return mask;
 919}
 920
 921/**
 922 *      relay_file_release - release file op for relay files
 923 *      @inode: the inode
 924 *      @filp: the file
 925 *
 926 *      Decrements the channel refcount, as the filesystem is
 927 *      no longer using it.
 928 */
 929static int relay_file_release(struct inode *inode, struct file *filp)
 930{
 931        struct rchan_buf *buf = filp->private_data;
 932        kref_put(&buf->kref, relay_remove_buf);
 933
 934        return 0;
 935}
 936
 937/*
 938 *      relay_file_read_consume - update the consumed count for the buffer
 939 */
 940static void relay_file_read_consume(struct rchan_buf *buf,
 941                                    size_t read_pos,
 942                                    size_t bytes_consumed)
 943{
 944        size_t subbuf_size = buf->chan->subbuf_size;
 945        size_t n_subbufs = buf->chan->n_subbufs;
 946        size_t read_subbuf;
 947
 948        if (buf->subbufs_produced == buf->subbufs_consumed &&
 949            buf->offset == buf->bytes_consumed)
 950                return;
 951
 952        if (buf->bytes_consumed + bytes_consumed > subbuf_size) {
 953                relay_subbufs_consumed(buf->chan, buf->cpu, 1);
 954                buf->bytes_consumed = 0;
 955        }
 956
 957        buf->bytes_consumed += bytes_consumed;
 958        if (!read_pos)
 959                read_subbuf = buf->subbufs_consumed % n_subbufs;
 960        else
 961                read_subbuf = read_pos / buf->chan->subbuf_size;
 962        if (buf->bytes_consumed + buf->padding[read_subbuf] == subbuf_size) {
 963                if ((read_subbuf == buf->subbufs_produced % n_subbufs) &&
 964                    (buf->offset == subbuf_size))
 965                        return;
 966                relay_subbufs_consumed(buf->chan, buf->cpu, 1);
 967                buf->bytes_consumed = 0;
 968        }
 969}
 970
 971/*
 972 *      relay_file_read_avail - boolean, are there unconsumed bytes available?
 973 */
 974static int relay_file_read_avail(struct rchan_buf *buf, size_t read_pos)
 975{
 976        size_t subbuf_size = buf->chan->subbuf_size;
 977        size_t n_subbufs = buf->chan->n_subbufs;
 978        size_t produced = buf->subbufs_produced;
 979        size_t consumed = buf->subbufs_consumed;
 980
 981        relay_file_read_consume(buf, read_pos, 0);
 982
 983        consumed = buf->subbufs_consumed;
 984
 985        if (unlikely(buf->offset > subbuf_size)) {
 986                if (produced == consumed)
 987                        return 0;
 988                return 1;
 989        }
 990
 991        if (unlikely(produced - consumed >= n_subbufs)) {
 992                consumed = produced - n_subbufs + 1;
 993                buf->subbufs_consumed = consumed;
 994                buf->bytes_consumed = 0;
 995        }
 996
 997        produced = (produced % n_subbufs) * subbuf_size + buf->offset;
 998        consumed = (consumed % n_subbufs) * subbuf_size + buf->bytes_consumed;
 999
1000        if (consumed > produced)
1001                produced += n_subbufs * subbuf_size;
1002
1003        if (consumed == produced) {
1004                if (buf->offset == subbuf_size &&
1005                    buf->subbufs_produced > buf->subbufs_consumed)
1006                        return 1;
1007                return 0;
1008        }
1009
1010        return 1;
1011}
1012
1013/**
1014 *      relay_file_read_subbuf_avail - return bytes available in sub-buffer
1015 *      @read_pos: file read position
1016 *      @buf: relay channel buffer
1017 */
1018static size_t relay_file_read_subbuf_avail(size_t read_pos,
1019                                           struct rchan_buf *buf)
1020{
1021        size_t padding, avail = 0;
1022        size_t read_subbuf, read_offset, write_subbuf, write_offset;
1023        size_t subbuf_size = buf->chan->subbuf_size;
1024
1025        write_subbuf = (buf->data - buf->start) / subbuf_size;
1026        write_offset = buf->offset > subbuf_size ? subbuf_size : buf->offset;
1027        read_subbuf = read_pos / subbuf_size;
1028        read_offset = read_pos % subbuf_size;
1029        padding = buf->padding[read_subbuf];
1030
1031        if (read_subbuf == write_subbuf) {
1032                if (read_offset + padding < write_offset)
1033                        avail = write_offset - (read_offset + padding);
1034        } else
1035                avail = (subbuf_size - padding) - read_offset;
1036
1037        return avail;
1038}
1039
1040/**
1041 *      relay_file_read_start_pos - find the first available byte to read
1042 *      @read_pos: file read position
1043 *      @buf: relay channel buffer
1044 *
1045 *      If the @read_pos is in the middle of padding, return the
1046 *      position of the first actually available byte, otherwise
1047 *      return the original value.
1048 */
1049static size_t relay_file_read_start_pos(size_t read_pos,
1050                                        struct rchan_buf *buf)
1051{
1052        size_t read_subbuf, padding, padding_start, padding_end;
1053        size_t subbuf_size = buf->chan->subbuf_size;
1054        size_t n_subbufs = buf->chan->n_subbufs;
1055        size_t consumed = buf->subbufs_consumed % n_subbufs;
1056
1057        if (!read_pos)
1058                read_pos = consumed * subbuf_size + buf->bytes_consumed;
1059        read_subbuf = read_pos / subbuf_size;
1060        padding = buf->padding[read_subbuf];
1061        padding_start = (read_subbuf + 1) * subbuf_size - padding;
1062        padding_end = (read_subbuf + 1) * subbuf_size;
1063        if (read_pos >= padding_start && read_pos < padding_end) {
1064                read_subbuf = (read_subbuf + 1) % n_subbufs;
1065                read_pos = read_subbuf * subbuf_size;
1066        }
1067
1068        return read_pos;
1069}
1070
1071/**
1072 *      relay_file_read_end_pos - return the new read position
1073 *      @read_pos: file read position
1074 *      @buf: relay channel buffer
1075 *      @count: number of bytes to be read
1076 */
1077static size_t relay_file_read_end_pos(struct rchan_buf *buf,
1078                                      size_t read_pos,
1079                                      size_t count)
1080{
1081        size_t read_subbuf, padding, end_pos;
1082        size_t subbuf_size = buf->chan->subbuf_size;
1083        size_t n_subbufs = buf->chan->n_subbufs;
1084
1085        read_subbuf = read_pos / subbuf_size;
1086        padding = buf->padding[read_subbuf];
1087        if (read_pos % subbuf_size + count + padding == subbuf_size)
1088                end_pos = (read_subbuf + 1) * subbuf_size;
1089        else
1090                end_pos = read_pos + count;
1091        if (end_pos >= subbuf_size * n_subbufs)
1092                end_pos = 0;
1093
1094        return end_pos;
1095}
1096
1097/*
1098 *      subbuf_read_actor - read up to one subbuf's worth of data
1099 */
1100static int subbuf_read_actor(size_t read_start,
1101                             struct rchan_buf *buf,
1102                             size_t avail,
1103                             read_descriptor_t *desc,
1104                             read_actor_t actor)
1105{
1106        void *from;
1107        int ret = 0;
1108
1109        from = buf->start + read_start;
1110        ret = avail;
1111        if (copy_to_user(desc->arg.buf, from, avail)) {
1112                desc->error = -EFAULT;
1113                ret = 0;
1114        }
1115        desc->arg.data += ret;
1116        desc->written += ret;
1117        desc->count -= ret;
1118
1119        return ret;
1120}
1121
1122typedef int (*subbuf_actor_t) (size_t read_start,
1123                               struct rchan_buf *buf,
1124                               size_t avail,
1125                               read_descriptor_t *desc,
1126                               read_actor_t actor);
1127
1128/*
1129 *      relay_file_read_subbufs - read count bytes, bridging subbuf boundaries
1130 */
1131static ssize_t relay_file_read_subbufs(struct file *filp, loff_t *ppos,
1132                                        subbuf_actor_t subbuf_actor,
1133                                        read_actor_t actor,
1134                                        read_descriptor_t *desc)
1135{
1136        struct rchan_buf *buf = filp->private_data;
1137        size_t read_start, avail;
1138        int ret;
1139
1140        if (!desc->count)
1141                return 0;
1142
1143        mutex_lock(&filp->f_path.dentry->d_inode->i_mutex);
1144        do {
1145                if (!relay_file_read_avail(buf, *ppos))
1146                        break;
1147
1148                read_start = relay_file_read_start_pos(*ppos, buf);
1149                avail = relay_file_read_subbuf_avail(read_start, buf);
1150                if (!avail)
1151                        break;
1152
1153                avail = min(desc->count, avail);
1154                ret = subbuf_actor(read_start, buf, avail, desc, actor);
1155                if (desc->error < 0)
1156                        break;
1157
1158                if (ret) {
1159                        relay_file_read_consume(buf, read_start, ret);
1160                        *ppos = relay_file_read_end_pos(buf, read_start, ret);
1161                }
1162        } while (desc->count && ret);
1163        mutex_unlock(&filp->f_path.dentry->d_inode->i_mutex);
1164
1165        return desc->written;
1166}
1167
1168static ssize_t relay_file_read(struct file *filp,
1169                               char __user *buffer,
1170                               size_t count,
1171                               loff_t *ppos)
1172{
1173        read_descriptor_t desc;
1174        desc.written = 0;
1175        desc.count = count;
1176        desc.arg.buf = buffer;
1177        desc.error = 0;
1178        return relay_file_read_subbufs(filp, ppos, subbuf_read_actor,
1179                                       NULL, &desc);
1180}
1181
1182static void relay_consume_bytes(struct rchan_buf *rbuf, int bytes_consumed)
1183{
1184        rbuf->bytes_consumed += bytes_consumed;
1185
1186        if (rbuf->bytes_consumed >= rbuf->chan->subbuf_size) {
1187                relay_subbufs_consumed(rbuf->chan, rbuf->cpu, 1);
1188                rbuf->bytes_consumed %= rbuf->chan->subbuf_size;
1189        }
1190}
1191
1192static void relay_pipe_buf_release(struct pipe_inode_info *pipe,
1193                                   struct pipe_buffer *buf)
1194{
1195        struct rchan_buf *rbuf;
1196
1197        rbuf = (struct rchan_buf *)page_private(buf->page);
1198        relay_consume_bytes(rbuf, buf->private);
1199}
1200
1201static struct pipe_buf_operations relay_pipe_buf_ops = {
1202        .can_merge = 0,
1203        .map = generic_pipe_buf_map,
1204        .unmap = generic_pipe_buf_unmap,
1205        .confirm = generic_pipe_buf_confirm,
1206        .release = relay_pipe_buf_release,
1207        .steal = generic_pipe_buf_steal,
1208        .get = generic_pipe_buf_get,
1209};
1210
1211static void relay_page_release(struct splice_pipe_desc *spd, unsigned int i)
1212{
1213}
1214
1215/*
1216 *      subbuf_splice_actor - splice up to one subbuf's worth of data
1217 */
1218static int subbuf_splice_actor(struct file *in,
1219                               loff_t *ppos,
1220                               struct pipe_inode_info *pipe,
1221                               size_t len,
1222                               unsigned int flags,
1223                               int *nonpad_ret)
1224{
1225        unsigned int pidx, poff, total_len, subbuf_pages, nr_pages, ret;
1226        struct rchan_buf *rbuf = in->private_data;
1227        unsigned int subbuf_size = rbuf->chan->subbuf_size;
1228        uint64_t pos = (uint64_t) *ppos;
1229        uint32_t alloc_size = (uint32_t) rbuf->chan->alloc_size;
1230        size_t read_start = (size_t) do_div(pos, alloc_size);
1231        size_t read_subbuf = read_start / subbuf_size;
1232        size_t padding = rbuf->padding[read_subbuf];
1233        size_t nonpad_end = read_subbuf * subbuf_size + subbuf_size - padding;
1234        struct page *pages[PIPE_BUFFERS];
1235        struct partial_page partial[PIPE_BUFFERS];
1236        struct splice_pipe_desc spd = {
1237                .pages = pages,
1238                .nr_pages = 0,
1239                .partial = partial,
1240                .flags = flags,
1241                .ops = &relay_pipe_buf_ops,
1242                .spd_release = relay_page_release,
1243        };
1244
1245        if (rbuf->subbufs_produced == rbuf->subbufs_consumed)
1246                return 0;
1247
1248        /*
1249         * Adjust read len, if longer than what is available
1250         */
1251        if (len > (subbuf_size - read_start % subbuf_size))
1252                len = subbuf_size - read_start % subbuf_size;
1253
1254        subbuf_pages = rbuf->chan->alloc_size >> PAGE_SHIFT;
1255        pidx = (read_start / PAGE_SIZE) % subbuf_pages;
1256        poff = read_start & ~PAGE_MASK;
1257        nr_pages = min_t(unsigned int, subbuf_pages, PIPE_BUFFERS);
1258
1259        for (total_len = 0; spd.nr_pages < nr_pages; spd.nr_pages++) {
1260                unsigned int this_len, this_end, private;
1261                unsigned int cur_pos = read_start + total_len;
1262
1263                if (!len)
1264                        break;
1265
1266                this_len = min_t(unsigned long, len, PAGE_SIZE - poff);
1267                private = this_len;
1268
1269                spd.pages[spd.nr_pages] = rbuf->page_array[pidx];
1270                spd.partial[spd.nr_pages].offset = poff;
1271
1272                this_end = cur_pos + this_len;
1273                if (this_end >= nonpad_end) {
1274                        this_len = nonpad_end - cur_pos;
1275                        private = this_len + padding;
1276                }
1277                spd.partial[spd.nr_pages].len = this_len;
1278                spd.partial[spd.nr_pages].private = private;
1279
1280                len -= this_len;
1281                total_len += this_len;
1282                poff = 0;
1283                pidx = (pidx + 1) % subbuf_pages;
1284
1285                if (this_end >= nonpad_end) {
1286                        spd.nr_pages++;
1287                        break;
1288                }
1289        }
1290
1291        if (!spd.nr_pages)
1292                return 0;
1293
1294        ret = *nonpad_ret = splice_to_pipe(pipe, &spd);
1295        if (ret < 0 || ret < total_len)
1296                return ret;
1297
1298        if (read_start + ret == nonpad_end)
1299                ret += padding;
1300
1301        return ret;
1302}
1303
1304static ssize_t relay_file_splice_read(struct file *in,
1305                                      loff_t *ppos,
1306                                      struct pipe_inode_info *pipe,
1307                                      size_t len,
1308                                      unsigned int flags)
1309{
1310        ssize_t spliced;
1311        int ret;
1312        int nonpad_ret = 0;
1313
1314        ret = 0;
1315        spliced = 0;
1316
1317        while (len && !spliced) {
1318                ret = subbuf_splice_actor(in, ppos, pipe, len, flags, &nonpad_ret);
1319                if (ret < 0)
1320                        break;
1321                else if (!ret) {
1322                        if (flags & SPLICE_F_NONBLOCK)
1323                                ret = -EAGAIN;
1324                        break;
1325                }
1326
1327                *ppos += ret;
1328                if (ret > len)
1329                        len = 0;
1330                else
1331                        len -= ret;
1332                spliced += nonpad_ret;
1333                nonpad_ret = 0;
1334        }
1335
1336        if (spliced)
1337                return spliced;
1338
1339        return ret;
1340}
1341
1342const struct file_operations relay_file_operations = {
1343        .open           = relay_file_open,
1344        .poll           = relay_file_poll,
1345        .mmap           = relay_file_mmap,
1346        .read           = relay_file_read,
1347        .llseek         = no_llseek,
1348        .release        = relay_file_release,
1349        .splice_read    = relay_file_splice_read,
1350};
1351EXPORT_SYMBOL_GPL(relay_file_operations);
1352
1353static __init int relay_init(void)
1354{
1355
1356        hotcpu_notifier(relay_hotcpu_callback, 0);
1357        return 0;
1358}
1359
1360early_initcall(relay_init);
1361