kernel/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/export.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 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  */
  71 static struct page **relay_alloc_page_array(unsigned int n_pages)
  72 {
  73         const size_t pa_size = n_pages * sizeof(struct page *);
  74         if (pa_size > PAGE_SIZE)
  75                 return vzalloc(pa_size);
  76         return kzalloc(pa_size, GFP_KERNEL);
  77 }
  78
  79 /*
  80  * free an array of pointers of struct page
  81  */
  82 static void relay_free_page_array(struct page **array)
  83 {
  84         if (is_vmalloc_addr(array))
  85                 vfree(array);
  86         else
  87                 kfree(array);
  88 }
  89
  90 /**
  91  *      relay_mmap_buf: - mmap channel buffer to process address space
  92  *      @buf: relay channel buffer
  93  *      @vma: vm_area_struct describing memory to be mapped
  94  *
  95  *      Returns 0 if ok, negative on error
  96  *
  97  *      Caller should already have grabbed mmap_sem.
  98  */
  99 static int relay_mmap_buf(struct rchan_buf *buf, struct vm_area_struct *vma)
 100 {
 101         unsigned long length = vma->vm_end - vma->vm_start;
 102         struct file *filp = vma->vm_file;
 103
 104         if (!buf)
 105                 return -EBADF;
 106
 107         if (length != (unsigned long)buf->chan->alloc_size)
 108                 return -EINVAL;
 109
 110         vma->vm_ops = &relay_file_mmap_ops;
 111         vma->vm_flags |= VM_DONTEXPAND;
 112         vma->vm_private_data = buf;
 113         buf->chan->cb->buf_mapped(buf, filp);
 114
 115         return 0;
 116 }
 117
 118 /**
 119  *      relay_alloc_buf - allocate a channel buffer
 120  *      @buf: the buffer struct
 121  *      @size: total size of the buffer
 122  *
 123  *      Returns a pointer to the resulting buffer, %NULL if unsuccessful. The
 124  *      passed in size will get page aligned, if it isn't already.
 125  */
 126 static void *relay_alloc_buf(struct rchan_buf *buf, size_t *size)
 127 {
 128         void *mem;
 129         unsigned int i, j, n_pages;
 130
 131         *size = PAGE_ALIGN(*size);
 132         n_pages = *size >> PAGE_SHIFT;
 133
 134         buf->page_array = relay_alloc_page_array(n_pages);
 135         if (!buf->page_array)
 136                 return NULL;
 137
 138         for (i = 0; i < n_pages; i++) {
 139                 buf->page_array[i] = alloc_page(GFP_KERNEL);
 140                 if (unlikely(!buf->page_array[i]))
 141                         goto depopulate;
 142                 set_page_private(buf->page_array[i], (unsigned long)buf);
 143         }
 144         mem = vmap(buf->page_array, n_pages, VM_MAP, PAGE_KERNEL);
 145         if (!mem)
 146                 goto depopulate;
 147
 148         memset(mem, 0, *size);
 149         buf->page_count = n_pages;
 150         return mem;
 151
 152 depopulate:
 153         for (j = 0; j < i; j++)
 154                 __free_page(buf->page_array[j]);
 155         relay_free_page_array(buf->page_array);
 156         return NULL;
 157 }
 158
 159 /**
 160  *      relay_create_buf - allocate and initialize a channel buffer
 161  *      @chan: the relay channel
 162  *
 163  *      Returns channel buffer if successful, %NULL otherwise.
 164  */
 165 static struct rchan_buf *relay_create_buf(struct rchan *chan)
 166 {
 167         struct rchan_buf *buf;
 168
 169         if (chan->n_subbufs > UINT_MAX / sizeof(size_t *))
 170                 return NULL;
 171
 172         buf = kzalloc(sizeof(struct rchan_buf), GFP_KERNEL);
 173         if (!buf)
 174                 return NULL;
 175         buf->padding = kmalloc(chan->n_subbufs * sizeof(size_t *), GFP_KERNEL);
 176         if (!buf->padding)
 177                 goto free_buf;
 178
 179         buf->start = relay_alloc_buf(buf, &chan->alloc_size);
 180         if (!buf->start)
 181                 goto free_buf;
 182
 183         buf->chan = chan;
 184         kref_get(&buf->chan->kref);
 185         return buf;
 186
 187 free_buf:
 188         kfree(buf->padding);
 189         kfree(buf);
 190         return NULL;
 191 }
 192
 193 /**
 194  *      relay_destroy_channel - free the channel struct
 195  *      @kref: target kernel reference that contains the relay channel
 196  *
 197  *      Should only be called from kref_put().
 198  */
 199 static void relay_destroy_channel(struct kref *kref)
 200 {
 201         struct rchan *chan = container_of(kref, struct rchan, kref);
 202         kfree(chan);
 203 }
 204
 205 /**
 206  *      relay_destroy_buf - destroy an rchan_buf struct and associated buffer
 207  *      @buf: the buffer struct
 208  */
 209 static void relay_destroy_buf(struct rchan_buf *buf)
 210 {
 211         struct rchan *chan = buf->chan;
 212         unsigned int i;
 213
 214         if (likely(buf->start)) {
 215                 vunmap(buf->start);
 216                 for (i = 0; i < buf->page_count; i++)
 217                         __free_page(buf->page_array[i]);
 218                 relay_free_page_array(buf->page_array);
 219         }
 220         chan->buf[buf->cpu] = NULL;
 221         kfree(buf->padding);
 222         kfree(buf);
 223         kref_put(&chan->kref, relay_destroy_channel);
 224 }
 225
 226 /**
 227  *      relay_remove_buf - remove a channel buffer
 228  *      @kref: target kernel reference that contains the relay buffer
 229  *
 230  *      Removes the file from the filesystem, which also frees the
 231  *      rchan_buf_struct and the channel buffer.  Should only be called from
 232  *      kref_put().
 233  */
 234 static void relay_remove_buf(struct kref *kref)
 235 {
 236         struct rchan_buf *buf = container_of(kref, struct rchan_buf, kref);
 237         relay_destroy_buf(buf);
 238 }
 239
 240 /**
 241  *      relay_buf_empty - boolean, is the channel buffer empty?
 242  *      @buf: channel buffer
 243  *
 244  *      Returns 1 if the buffer is empty, 0 otherwise.
 245  */
 246 static int relay_buf_empty(struct rchan_buf *buf)
 247 {
 248         return (buf->subbufs_produced - buf->subbufs_consumed) ? 0 : 1;
 249 }
 250
 251 /**
 252  *      relay_buf_full - boolean, is the channel buffer full?
 253  *      @buf: channel buffer
 254  *
 255  *      Returns 1 if the buffer is full, 0 otherwise.
 256  */
 257 int relay_buf_full(struct rchan_buf *buf)
 258 {
 259         size_t ready = buf->subbufs_produced - buf->subbufs_consumed;
 260         return (ready >= buf->chan->n_subbufs) ? 1 : 0;
 261 }
 262 EXPORT_SYMBOL_GPL(relay_buf_full);
 263
 264 /*
 265  * High-level relay kernel API and associated functions.
 266  */
 267
 268 /*
 269  * rchan_callback implementations defining default channel behavior.  Used
 270  * in place of corresponding NULL values in client callback struct.
 271  */
 272
 273 /*
 274  * subbuf_start() default callback.  Does nothing.
 275  */
 276 static int subbuf_start_default_callback (struct rchan_buf *buf,
 277                                           void *subbuf,
 278                                           void *prev_subbuf,
 279                                           size_t prev_padding)
 280 {
 281         if (relay_buf_full(buf))
 282                 return 0;
 283
 284         return 1;
 285 }
 286
 287 /*
 288  * buf_mapped() default callback.  Does nothing.
 289  */
 290 static void buf_mapped_default_callback(struct rchan_buf *buf,
 291                                         struct file *filp)
 292 {
 293 }
 294
 295 /*
 296  * buf_unmapped() default callback.  Does nothing.
 297  */
 298 static void buf_unmapped_default_callback(struct rchan_buf *buf,
 299                                           struct file *filp)
 300 {
 301 }
 302
 303 /*
 304  * create_buf_file_create() default callback.  Does nothing.
 305  */
 306 static struct dentry *create_buf_file_default_callback(const char *filename,
 307                                                        struct dentry *parent,
 308                                                        umode_t mode,
 309                                                        struct rchan_buf *buf,
 310                                                        int *is_global)
 311 {
 312         return NULL;
 313 }
 314
 315 /*
 316  * remove_buf_file() default callback.  Does nothing.
 317  */
 318 static int remove_buf_file_default_callback(struct dentry *dentry)
 319 {
 320         return -EINVAL;
 321 }
 322
 323 /* relay channel default callbacks */
 324 static struct rchan_callbacks default_channel_callbacks = {
 325         .subbuf_start = subbuf_start_default_callback,
 326         .buf_mapped = buf_mapped_default_callback,
 327         .buf_unmapped = buf_unmapped_default_callback,
 328         .create_buf_file = create_buf_file_default_callback,
 329         .remove_buf_file = remove_buf_file_default_callback,
 330 };
 331
 332 /**
 333  *      wakeup_readers - wake up readers waiting on a channel
 334  *      @data: contains the channel buffer
 335  *
 336  *      This is the timer function used to defer reader waking.
 337  */
 338 static void wakeup_readers(unsigned long data)
 339 {
 340         struct rchan_buf *buf = (struct rchan_buf *)data;
 341         wake_up_interruptible(&buf->read_wait);
 342         /*
 343          * Stupid polling for now:
 344          */
 345         mod_timer(&buf->timer, jiffies + 1);
 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                 mod_timer(&buf->timer, jiffies + 1);
 364         } else
 365                 del_timer_sync(&buf->timer);
 366
 367         buf->subbufs_produced = 0;
 368         buf->subbufs_consumed = 0;
 369         buf->bytes_consumed = 0;
 370         buf->finalized = 0;
 371         buf->data = buf->start;
 372         buf->offset = 0;
 373
 374         for (i = 0; i < buf->chan->n_subbufs; i++)
 375                 buf->padding[i] = 0;
 376
 377         buf->chan->cb->subbuf_start(buf, buf->data, NULL, 0);
 378 }
 379
 380 /**
 381  *      relay_reset - reset the channel
 382  *      @chan: the channel
 383  *
 384  *      This has the effect of erasing all data from all channel buffers
 385  *      and restarting the channel in its initial state.  The buffers
 386  *      are not freed, so any mappings are still in effect.
 387  *
 388  *      NOTE. Care should be taken that the channel isn't actually
 389  *      being used by anything when this call is made.
 390  */
 391 void relay_reset(struct rchan *chan)
 392 {
 393         unsigned int i;
 394
 395         if (!chan)
 396                 return;
 397
 398         if (chan->is_global && chan->buf[0]) {
 399                 __relay_reset(chan->buf[0], 0);
 400                 return;
 401         }
 402
 403         mutex_lock(&relay_channels_mutex);
 404         for_each_possible_cpu(i)
 405                 if (chan->buf[i])
 406                         __relay_reset(chan->buf[i], 0);
 407         mutex_unlock(&relay_channels_mutex);
 408 }
 409 EXPORT_SYMBOL_GPL(relay_reset);
 410
 411 static inline void relay_set_buf_dentry(struct rchan_buf *buf,
 412                                         struct dentry *dentry)
 413 {
 414         buf->dentry = dentry;
 415         d_inode(buf->dentry)->i_size = buf->early_bytes;
 416 }
 417
 418 static struct dentry *relay_create_buf_file(struct rchan *chan,
 419                                             struct rchan_buf *buf,
 420                                             unsigned int cpu)
 421 {
 422         struct dentry *dentry;
 423         char *tmpname;
 424
 425         tmpname = kzalloc(NAME_MAX + 1, GFP_KERNEL);
 426         if (!tmpname)
 427                 return NULL;
 428         snprintf(tmpname, NAME_MAX, "%s%d", chan->base_filename, cpu);
 429
 430         /* Create file in fs */
 431         dentry = chan->cb->create_buf_file(tmpname, chan->parent,
 432                                            S_IRUSR, buf,
 433                                            &chan->is_global);
 434
 435         kfree(tmpname);
 436
 437         return dentry;
 438 }
 439
 440 /*
 441  *      relay_open_buf - create a new relay channel buffer
 442  *
 443  *      used by relay_open() and CPU hotplug.
 444  */
 445 static struct rchan_buf *relay_open_buf(struct rchan *chan, unsigned int cpu)
 446 {
 447         struct rchan_buf *buf = NULL;
 448         struct dentry *dentry;
 449
 450         if (chan->is_global)
 451                 return chan->buf[0];
 452
 453         buf = relay_create_buf(chan);
 454         if (!buf)
 455                 return NULL;
 456
 457         if (chan->has_base_filename) {
 458                 dentry = relay_create_buf_file(chan, buf, cpu);
 459                 if (!dentry)
 460                         goto free_buf;
 461                 relay_set_buf_dentry(buf, dentry);
 462         }
 463
 464         buf->cpu = cpu;
 465         __relay_reset(buf, 1);
 466
 467         if(chan->is_global) {
 468                 chan->buf[0] = buf;
 469                 buf->cpu = 0;
 470         }
 471
 472         return buf;
 473
 474 free_buf:
 475         relay_destroy_buf(buf);
 476         return NULL;
 477 }
 478
 479 /**
 480  *      relay_close_buf - close a channel buffer
 481  *      @buf: channel buffer
 482  *
 483  *      Marks the buffer finalized and restores the default callbacks.
 484  *      The channel buffer and channel buffer data structure are then freed
 485  *      automatically when the last reference is given up.
 486  */
 487 static void relay_close_buf(struct rchan_buf *buf)
 488 {
 489         buf->finalized = 1;
 490         del_timer_sync(&buf->timer);
 491         buf->chan->cb->remove_buf_file(buf->dentry);
 492         kref_put(&buf->kref, relay_remove_buf);
 493 }
 494
 495 static void setup_callbacks(struct rchan *chan,
 496                                    struct rchan_callbacks *cb)
 497 {
 498         if (!cb) {
 499                 chan->cb = &default_channel_callbacks;
 500                 return;
 501         }
 502
 503         if (!cb->subbuf_start)
 504                 cb->subbuf_start = subbuf_start_default_callback;
 505         if (!cb->buf_mapped)
 506                 cb->buf_mapped = buf_mapped_default_callback;
 507         if (!cb->buf_unmapped)
 508                 cb->buf_unmapped = buf_unmapped_default_callback;
 509         if (!cb->create_buf_file)
 510                 cb->create_buf_file = create_buf_file_default_callback;
 511         if (!cb->remove_buf_file)
 512                 cb->remove_buf_file = remove_buf_file_default_callback;
 513         chan->cb = cb;
 514 }
 515
 516 /**
 517  *      relay_hotcpu_callback - CPU hotplug callback
 518  *      @nb: notifier block
 519  *      @action: hotplug action to take
 520  *      @hcpu: CPU number
 521  *
 522  *      Returns the success/failure of the operation. (%NOTIFY_OK, %NOTIFY_BAD)
 523  */
 524 static int relay_hotcpu_callback(struct notifier_block *nb,
 525                                 unsigned long action,
 526                                 void *hcpu)
 527 {
 528         unsigned int hotcpu = (unsigned long)hcpu;
 529         struct rchan *chan;
 530
 531         switch(action) {
 532         case CPU_UP_PREPARE:
 533         case CPU_UP_PREPARE_FROZEN:
 534                 mutex_lock(&relay_channels_mutex);
 535                 list_for_each_entry(chan, &relay_channels, list) {
 536                         if (chan->buf[hotcpu])
 537                                 continue;
 538                         chan->buf[hotcpu] = relay_open_buf(chan, hotcpu);
 539                         if(!chan->buf[hotcpu]) {
 540                                 printk(KERN_ERR
 541                                         "relay_hotcpu_callback: cpu %d buffer "
 542                                         "creation failed\n", hotcpu);
 543                                 mutex_unlock(&relay_channels_mutex);
 544                                 return notifier_from_errno(-ENOMEM);
 545                         }
 546                 }
 547                 mutex_unlock(&relay_channels_mutex);
 548                 break;
 549         case CPU_DEAD:
 550         case CPU_DEAD_FROZEN:
 551                 /* No need to flush the cpu : will be flushed upon
 552                  * final relay_flush() call. */
 553                 break;
 554         }
 555         return NOTIFY_OK;
 556 }
 557
 558 /**
 559  *      relay_open - create a new relay channel
 560  *      @base_filename: base name of files to create, %NULL for buffering only
 561  *      @parent: dentry of parent directory, %NULL for root directory or buffer
 562  *      @subbuf_size: size of sub-buffers
 563  *      @n_subbufs: number of sub-buffers
 564  *      @cb: client callback functions
 565  *      @private_data: user-defined data
 566  *
 567  *      Returns channel pointer if successful, %NULL otherwise.
 568  *
 569  *      Creates a channel buffer for each cpu using the sizes and
 570  *      attributes specified.  The created channel buffer files
 571  *      will be named base_filename0...base_filenameN-1.  File
 572  *      permissions will be %S_IRUSR.
 573  */
 574 struct rchan *relay_open(const char *base_filename,
 575                          struct dentry *parent,
 576                          size_t subbuf_size,
 577                          size_t n_subbufs,
 578                          struct rchan_callbacks *cb,
 579                          void *private_data)
 580 {
 581         unsigned int i;
 582         struct rchan *chan;
 583
 584         if (!(subbuf_size && n_subbufs))
 585                 return NULL;
 586         if (subbuf_size > UINT_MAX / n_subbufs)
 587                 return NULL;
 588
 589         chan = kzalloc(sizeof(struct rchan), GFP_KERNEL);
 590         if (!chan)
 591                 return NULL;
 592
 593         chan->version = RELAYFS_CHANNEL_VERSION;
 594         chan->n_subbufs = n_subbufs;
 595         chan->subbuf_size = subbuf_size;
 596         chan->alloc_size = PAGE_ALIGN(subbuf_size * n_subbufs);
 597         chan->parent = parent;
 598         chan->private_data = private_data;
 599         if (base_filename) {
 600                 chan->has_base_filename = 1;
 601                 strlcpy(chan->base_filename, base_filename, NAME_MAX);
 602         }
 603         setup_callbacks(chan, cb);
 604         kref_init(&chan->kref);
 605
 606         mutex_lock(&relay_channels_mutex);
 607         for_each_online_cpu(i) {
 608                 chan->buf[i] = relay_open_buf(chan, i);
 609                 if (!chan->buf[i])
 610                         goto free_bufs;
 611         }
 612         list_add(&chan->list, &relay_channels);
 613         mutex_unlock(&relay_channels_mutex);
 614
 615         return chan;
 616
 617 free_bufs:
 618         for_each_possible_cpu(i) {
 619                 if (chan->buf[i])
 620                         relay_close_buf(chan->buf[i]);
 621         }
 622
 623         kref_put(&chan->kref, relay_destroy_channel);
 624         mutex_unlock(&relay_channels_mutex);
 625         return NULL;
 626 }
 627 EXPORT_SYMBOL_GPL(relay_open);
 628
 629 struct rchan_percpu_buf_dispatcher {
 630         struct rchan_buf *buf;
 631         struct dentry *dentry;
 632 };
 633
 634 /* Called in atomic context. */
 635 static void __relay_set_buf_dentry(void *info)
 636 {
 637         struct rchan_percpu_buf_dispatcher *p = info;
 638
 639         relay_set_buf_dentry(p->buf, p->dentry);
 640 }
 641
 642 /**
 643  *      relay_late_setup_files - triggers file creation
 644  *      @chan: channel to operate on
 645  *      @base_filename: base name of files to create
 646  *      @parent: dentry of parent directory, %NULL for root directory
 647  *
 648  *      Returns 0 if successful, non-zero otherwise.
 649  *
 650  *      Use to setup files for a previously buffer-only channel.
 651  *      Useful to do early tracing in kernel, before VFS is up, for example.
 652  */
 653 int relay_late_setup_files(struct rchan *chan,
 654                            const char *base_filename,
 655                            struct dentry *parent)
 656 {
 657         int err = 0;
 658         unsigned int i, curr_cpu;
 659         unsigned long flags;
 660         struct dentry *dentry;
 661         struct rchan_percpu_buf_dispatcher disp;
 662
 663         if (!chan || !base_filename)
 664                 return -EINVAL;
 665
 666         strlcpy(chan->base_filename, base_filename, NAME_MAX);
 667
 668         mutex_lock(&relay_channels_mutex);
 669         /* Is chan already set up? */
 670         if (unlikely(chan->has_base_filename)) {
 671                 mutex_unlock(&relay_channels_mutex);
 672                 return -EEXIST;
 673         }
 674         chan->has_base_filename = 1;
 675         chan->parent = parent;
 676         curr_cpu = get_cpu();
 677         /*
 678          * The CPU hotplug notifier ran before us and created buffers with
 679          * no files associated. So it's safe to call relay_setup_buf_file()
 680          * on all currently online CPUs.
 681          */
 682         for_each_online_cpu(i) {
 683                 if (unlikely(!chan->buf[i])) {
 684                         WARN_ONCE(1, KERN_ERR "CPU has no buffer!\n");
 685                         err = -EINVAL;
 686                         break;
 687                 }
 688
 689                 dentry = relay_create_buf_file(chan, chan->buf[i], i);
 690                 if (unlikely(!dentry)) {
 691                         err = -EINVAL;
 692                         break;
 693                 }
 694
 695                 if (curr_cpu == i) {
 696                         local_irq_save(flags);
 697                         relay_set_buf_dentry(chan->buf[i], dentry);
 698                         local_irq_restore(flags);
 699                 } else {
 700                         disp.buf = chan->buf[i];
 701                         disp.dentry = dentry;
 702                         smp_mb();
 703                         /* relay_channels_mutex must be held, so wait. */
 704                         err = smp_call_function_single(i,
 705                                                        __relay_set_buf_dentry,
 706                                                        &disp, 1);
 707                 }
 708                 if (unlikely(err))
 709                         break;
 710         }
 711         put_cpu();
 712         mutex_unlock(&relay_channels_mutex);
 713
 714         return err;
 715 }
 716
 717 /**
 718  *      relay_switch_subbuf - switch to a new sub-buffer
 719  *      @buf: channel buffer
 720  *      @length: size of current event
 721  *
 722  *      Returns either the length passed in or 0 if full.
 723  *
 724  *      Performs sub-buffer-switch tasks such as invoking callbacks,
 725  *      updating padding counts, waking up readers, etc.
 726  */
 727 size_t relay_switch_subbuf(struct rchan_buf *buf, size_t length)
 728 {
 729         void *old, *new;
 730         size_t old_subbuf, new_subbuf;
 731
 732         if (unlikely(length > buf->chan->subbuf_size))
 733                 goto toobig;
 734
 735         if (buf->offset != buf->chan->subbuf_size + 1) {
 736                 buf->prev_padding = buf->chan->subbuf_size - buf->offset;
 737                 old_subbuf = buf->subbufs_produced % buf->chan->n_subbufs;
 738                 buf->padding[old_subbuf] = buf->prev_padding;
 739                 buf->subbufs_produced++;
 740                 if (buf->dentry)
 741                         d_inode(buf->dentry)->i_size +=
 742                                 buf->chan->subbuf_size -
 743                                 buf->padding[old_subbuf];
 744                 else
 745                         buf->early_bytes += buf->chan->subbuf_size -
 746                                             buf->padding[old_subbuf];
 747         }
 748
 749         old = buf->data;
 750         new_subbuf = buf->subbufs_produced % buf->chan->n_subbufs;
 751         new = buf->start + new_subbuf * buf->chan->subbuf_size;
 752         buf->offset = 0;
 753         if (!buf->chan->cb->subbuf_start(buf, new, old, buf->prev_padding)) {
 754                 buf->offset = buf->chan->subbuf_size + 1;
 755                 return 0;
 756         }
 757         buf->data = new;
 758         buf->padding[new_subbuf] = 0;
 759
 760         if (unlikely(length + buf->offset > buf->chan->subbuf_size))
 761                 goto toobig;
 762
 763         return length;
 764
 765 toobig:
 766         buf->chan->last_toobig = length;
 767         return 0;
 768 }
 769 EXPORT_SYMBOL_GPL(relay_switch_subbuf);
 770
 771 /**
 772  *      relay_subbufs_consumed - update the buffer's sub-buffers-consumed count
 773  *      @chan: the channel
 774  *      @cpu: the cpu associated with the channel buffer to update
 775  *      @subbufs_consumed: number of sub-buffers to add to current buf's count
 776  *
 777  *      Adds to the channel buffer's consumed sub-buffer count.
 778  *      subbufs_consumed should be the number of sub-buffers newly consumed,
 779  *      not the total consumed.
 780  *
 781  *      NOTE. Kernel clients don't need to call this function if the channel
 782  *      mode is 'overwrite'.
 783  */
 784 void relay_subbufs_consumed(struct rchan *chan,
 785                             unsigned int cpu,
 786                             size_t subbufs_consumed)
 787 {
 788         struct rchan_buf *buf;
 789
 790         if (!chan)
 791                 return;
 792
 793         if (cpu >= NR_CPUS || !chan->buf[cpu] ||
 794                                         subbufs_consumed > chan->n_subbufs)
 795                 return;
 796
 797         buf = chan->buf[cpu];
 798         if (subbufs_consumed > buf->subbufs_produced - buf->subbufs_consumed)
 799                 buf->subbufs_consumed = buf->subbufs_produced;
 800         else
 801                 buf->subbufs_consumed += subbufs_consumed;
 802 }
 803 EXPORT_SYMBOL_GPL(relay_subbufs_consumed);
 804
 805 /**
 806  *      relay_close - close the channel
 807  *      @chan: the channel
 808  *
 809  *      Closes all channel buffers and frees the channel.
 810  */
 811 void relay_close(struct rchan *chan)
 812 {
 813         unsigned int i;
 814
 815         if (!chan)
 816                 return;
 817
 818         mutex_lock(&relay_channels_mutex);
 819         if (chan->is_global && chan->buf[0])
 820                 relay_close_buf(chan->buf[0]);
 821         else
 822                 for_each_possible_cpu(i)
 823                         if (chan->buf[i])
 824                                 relay_close_buf(chan->buf[i]);
 825
 826         if (chan->last_toobig)
 827                 printk(KERN_WARNING "relay: one or more items not logged "
 828                        "[item size (%Zd) > sub-buffer size (%Zd)]\n",
 829                        chan->last_toobig, chan->subbuf_size);
 830
 831         list_del(&chan->list);
 832         kref_put(&chan->kref, relay_destroy_channel);
 833         mutex_unlock(&relay_channels_mutex);
 834 }
 835 EXPORT_SYMBOL_GPL(relay_close);
 836
 837 /**
 838  *      relay_flush - close the channel
 839  *      @chan: the channel
 840  *
 841  *      Flushes all channel buffers, i.e. forces buffer switch.
 842  */
 843 void relay_flush(struct rchan *chan)
 844 {
 845         unsigned int i;
 846
 847         if (!chan)
 848                 return;
 849
 850         if (chan->is_global && chan->buf[0]) {
 851                 relay_switch_subbuf(chan->buf[0], 0);
 852                 return;
 853         }
 854
 855         mutex_lock(&relay_channels_mutex);
 856         for_each_possible_cpu(i)
 857                 if (chan->buf[i])
 858                         relay_switch_subbuf(chan->buf[i], 0);
 859         mutex_unlock(&relay_channels_mutex);
 860 }
 861 EXPORT_SYMBOL_GPL(relay_flush);
 862
 863 /**
 864  *      relay_file_open - open file op for relay files
 865  *      @inode: the inode
 866  *      @filp: the file
 867  *
 868  *      Increments the channel buffer refcount.
 869  */
 870 static int relay_file_open(struct inode *inode, struct file *filp)
 871 {
 872         struct rchan_buf *buf = inode->i_private;
 873         kref_get(&buf->kref);
 874         filp->private_data = buf;
 875
 876         return nonseekable_open(inode, filp);
 877 }
 878
 879 /**
 880  *      relay_file_mmap - mmap file op for relay files
 881  *      @filp: the file
 882  *      @vma: the vma describing what to map
 883  *
 884  *      Calls upon relay_mmap_buf() to map the file into user space.
 885  */
 886 static int relay_file_mmap(struct file *filp, struct vm_area_struct *vma)
 887 {
 888         struct rchan_buf *buf = filp->private_data;
 889         return relay_mmap_buf(buf, vma);
 890 }
 891
 892 /**
 893  *      relay_file_poll - poll file op for relay files
 894  *      @filp: the file
 895  *      @wait: poll table
 896  *
 897  *      Poll implemention.
 898  */
 899 static unsigned int relay_file_poll(struct file *filp, poll_table *wait)
 900 {
 901         unsigned int mask = 0;
 902         struct rchan_buf *buf = filp->private_data;
 903
 904         if (buf->finalized)
 905                 return POLLERR;
 906
 907         if (filp->f_mode & FMODE_READ) {
 908                 poll_wait(filp, &buf->read_wait, wait);
 909                 if (!relay_buf_empty(buf))
 910                         mask |= POLLIN | POLLRDNORM;
 911         }
 912
 913         return mask;
 914 }
 915
 916 /**
 917  *      relay_file_release - release file op for relay files
 918  *      @inode: the inode
 919  *      @filp: the file
 920  *
 921  *      Decrements the channel refcount, as the filesystem is
 922  *      no longer using it.
 923  */
 924 static int relay_file_release(struct inode *inode, struct file *filp)
 925 {
 926         struct rchan_buf *buf = filp->private_data;
 927         kref_put(&buf->kref, relay_remove_buf);
 928
 929         return 0;
 930 }
 931
 932 /*
 933  *      relay_file_read_consume - update the consumed count for the buffer
 934  */
 935 static void relay_file_read_consume(struct rchan_buf *buf,
 936                                     size_t read_pos,
 937                                     size_t bytes_consumed)
 938 {
 939         size_t subbuf_size = buf->chan->subbuf_size;
 940         size_t n_subbufs = buf->chan->n_subbufs;
 941         size_t read_subbuf;
 942
 943         if (buf->subbufs_produced == buf->subbufs_consumed &&
 944             buf->offset == buf->bytes_consumed)
 945                 return;
 946
 947         if (buf->bytes_consumed + bytes_consumed > subbuf_size) {
 948                 relay_subbufs_consumed(buf->chan, buf->cpu, 1);
 949                 buf->bytes_consumed = 0;
 950         }
 951
 952         buf->bytes_consumed += bytes_consumed;
 953         if (!read_pos)
 954                 read_subbuf = buf->subbufs_consumed % n_subbufs;
 955         else
 956                 read_subbuf = read_pos / buf->chan->subbuf_size;
 957         if (buf->bytes_consumed + buf->padding[read_subbuf] == subbuf_size) {
 958                 if ((read_subbuf == buf->subbufs_produced % n_subbufs) &&
 959                     (buf->offset == subbuf_size))
 960                         return;
 961                 relay_subbufs_consumed(buf->chan, buf->cpu, 1);
 962                 buf->bytes_consumed = 0;
 963         }
 964 }
 965
 966 /*
 967  *      relay_file_read_avail - boolean, are there unconsumed bytes available?
 968  */
 969 static int relay_file_read_avail(struct rchan_buf *buf, size_t read_pos)
 970 {
 971         size_t subbuf_size = buf->chan->subbuf_size;
 972         size_t n_subbufs = buf->chan->n_subbufs;
 973         size_t produced = buf->subbufs_produced;
 974         size_t consumed = buf->subbufs_consumed;
 975
 976         relay_file_read_consume(buf, read_pos, 0);
 977
 978         consumed = buf->subbufs_consumed;
 979
 980         if (unlikely(buf->offset > subbuf_size)) {
 981                 if (produced == consumed)
 982                         return 0;
 983                 return 1;
 984         }
 985
 986         if (unlikely(produced - consumed >= n_subbufs)) {
 987                 consumed = produced - n_subbufs + 1;
 988                 buf->subbufs_consumed = consumed;
 989                 buf->bytes_consumed = 0;
 990         }
 991
 992         produced = (produced % n_subbufs) * subbuf_size + buf->offset;
 993         consumed = (consumed % n_subbufs) * subbuf_size + buf->bytes_consumed;
 994
 995         if (consumed > produced)
 996                 produced += n_subbufs * subbuf_size;
 997
 998         if (consumed == produced) {
 999                 if (buf->offset == subbuf_size &&
1000                     buf->subbufs_produced > buf->subbufs_consumed)
1001                         return 1;
1002                 return 0;
1003         }
1004
1005         return 1;
1006 }
1007
1008 /**
1009  *      relay_file_read_subbuf_avail - return bytes available in sub-buffer
1010  *      @read_pos: file read position
1011  *      @buf: relay channel buffer
1012  */
1013 static size_t relay_file_read_subbuf_avail(size_t read_pos,
1014                                            struct rchan_buf *buf)
1015 {
1016         size_t padding, avail = 0;
1017         size_t read_subbuf, read_offset, write_subbuf, write_offset;
1018         size_t subbuf_size = buf->chan->subbuf_size;
1019
1020         write_subbuf = (buf->data - buf->start) / subbuf_size;
1021         write_offset = buf->offset > subbuf_size ? subbuf_size : buf->offset;
1022         read_subbuf = read_pos / subbuf_size;
1023         read_offset = read_pos % subbuf_size;
1024         padding = buf->padding[read_subbuf];
1025
1026         if (read_subbuf == write_subbuf) {
1027                 if (read_offset + padding < write_offset)
1028                         avail = write_offset - (read_offset + padding);
1029         } else
1030                 avail = (subbuf_size - padding) - read_offset;
1031
1032         return avail;
1033 }
1034
1035 /**
1036  *      relay_file_read_start_pos - find the first available byte to read
1037  *      @read_pos: file read position
1038  *      @buf: relay channel buffer
1039  *
1040  *      If the @read_pos is in the middle of padding, return the
1041  *      position of the first actually available byte, otherwise
1042  *      return the original value.
1043  */
1044 static size_t relay_file_read_start_pos(size_t read_pos,
1045                                         struct rchan_buf *buf)
1046 {
1047         size_t read_subbuf, padding, padding_start, padding_end;
1048         size_t subbuf_size = buf->chan->subbuf_size;
1049         size_t n_subbufs = buf->chan->n_subbufs;
1050         size_t consumed = buf->subbufs_consumed % n_subbufs;
1051
1052         if (!read_pos)
1053                 read_pos = consumed * subbuf_size + buf->bytes_consumed;
1054         read_subbuf = read_pos / subbuf_size;
1055         padding = buf->padding[read_subbuf];
1056         padding_start = (read_subbuf + 1) * subbuf_size - padding;
1057         padding_end = (read_subbuf + 1) * subbuf_size;
1058         if (read_pos >= padding_start && read_pos < padding_end) {
1059                 read_subbuf = (read_subbuf + 1) % n_subbufs;
1060                 read_pos = read_subbuf * subbuf_size;
1061         }
1062
1063         return read_pos;
1064 }
1065
1066 /**
1067  *      relay_file_read_end_pos - return the new read position
1068  *      @read_pos: file read position
1069  *      @buf: relay channel buffer
1070  *      @count: number of bytes to be read
1071  */
1072 static size_t relay_file_read_end_pos(struct rchan_buf *buf,
1073                                       size_t read_pos,
1074                                       size_t count)
1075 {
1076         size_t read_subbuf, padding, end_pos;
1077         size_t subbuf_size = buf->chan->subbuf_size;
1078         size_t n_subbufs = buf->chan->n_subbufs;
1079
1080         read_subbuf = read_pos / subbuf_size;
1081         padding = buf->padding[read_subbuf];
1082         if (read_pos % subbuf_size + count + padding == subbuf_size)
1083                 end_pos = (read_subbuf + 1) * subbuf_size;
1084         else
1085                 end_pos = read_pos + count;
1086         if (end_pos >= subbuf_size * n_subbufs)
1087                 end_pos = 0;
1088
1089         return end_pos;
1090 }
1091
1092 /*
1093  *      subbuf_read_actor - read up to one subbuf's worth of data
1094  */
1095 static int subbuf_read_actor(size_t read_start,
1096                              struct rchan_buf *buf,
1097                              size_t avail,
1098                              read_descriptor_t *desc)
1099 {
1100         void *from;
1101         int ret = 0;
1102
1103         from = buf->start + read_start;
1104         ret = avail;
1105         if (copy_to_user(desc->arg.buf, from, avail)) {
1106                 desc->error = -EFAULT;
1107                 ret = 0;
1108         }
1109         desc->arg.data += ret;
1110         desc->written += ret;
1111         desc->count -= ret;
1112
1113         return ret;
1114 }
1115
1116 typedef int (*subbuf_actor_t) (size_t read_start,
1117                                struct rchan_buf *buf,
1118                                size_t avail,
1119                                read_descriptor_t *desc);
1120
1121 /*
1122  *      relay_file_read_subbufs - read count bytes, bridging subbuf boundaries
1123  */
1124 static ssize_t relay_file_read_subbufs(struct file *filp, loff_t *ppos,
1125                                         subbuf_actor_t subbuf_actor,
1126                                         read_descriptor_t *desc)
1127 {
1128         struct rchan_buf *buf = filp->private_data;
1129         size_t read_start, avail;
1130         int ret;
1131
1132         if (!desc->count)
1133                 return 0;
1134
1135         mutex_lock(&file_inode(filp)->i_mutex);
1136         do {
1137                 if (!relay_file_read_avail(buf, *ppos))
1138                         break;
1139
1140                 read_start = relay_file_read_start_pos(*ppos, buf);
1141                 avail = relay_file_read_subbuf_avail(read_start, buf);
1142                 if (!avail)
1143                         break;
1144
1145                 avail = min(desc->count, avail);
1146                 ret = subbuf_actor(read_start, buf, avail, desc);
1147                 if (desc->error < 0)
1148                         break;
1149
1150                 if (ret) {
1151                         relay_file_read_consume(buf, read_start, ret);
1152                         *ppos = relay_file_read_end_pos(buf, read_start, ret);
1153                 }
1154         } while (desc->count && ret);
1155         mutex_unlock(&file_inode(filp)->i_mutex);
1156
1157         return desc->written;
1158 }
1159
1160 static ssize_t relay_file_read(struct file *filp,
1161                                char __user *buffer,
1162                                size_t count,
1163                                loff_t *ppos)
1164 {
1165         read_descriptor_t desc;
1166         desc.written = 0;
1167         desc.count = count;
1168         desc.arg.buf = buffer;
1169         desc.error = 0;
1170         return relay_file_read_subbufs(filp, ppos, subbuf_read_actor, &desc);
1171 }
1172
1173 static void relay_consume_bytes(struct rchan_buf *rbuf, int bytes_consumed)
1174 {
1175         rbuf->bytes_consumed += bytes_consumed;
1176
1177         if (rbuf->bytes_consumed >= rbuf->chan->subbuf_size) {
1178                 relay_subbufs_consumed(rbuf->chan, rbuf->cpu, 1);
1179                 rbuf->bytes_consumed %= rbuf->chan->subbuf_size;
1180         }
1181 }
1182
1183 static void relay_pipe_buf_release(struct pipe_inode_info *pipe,
1184                                    struct pipe_buffer *buf)
1185 {
1186         struct rchan_buf *rbuf;
1187
1188         rbuf = (struct rchan_buf *)page_private(buf->page);
1189         relay_consume_bytes(rbuf, buf->private);
1190 }
1191
1192 static const struct pipe_buf_operations relay_pipe_buf_ops = {
1193         .can_merge = 0,
1194         .confirm = generic_pipe_buf_confirm,
1195         .release = relay_pipe_buf_release,
1196         .steal = generic_pipe_buf_steal,
1197         .get = generic_pipe_buf_get,
1198 };
1199
1200 static void relay_page_release(struct splice_pipe_desc *spd, unsigned int i)
1201 {
1202 }
1203
1204 /*
1205  *      subbuf_splice_actor - splice up to one subbuf's worth of data
1206  */
1207 static ssize_t subbuf_splice_actor(struct file *in,
1208                                loff_t *ppos,
1209                                struct pipe_inode_info *pipe,
1210                                size_t len,
1211                                unsigned int flags,
1212                                int *nonpad_ret)
1213 {
1214         unsigned int pidx, poff, total_len, subbuf_pages, nr_pages;
1215         struct rchan_buf *rbuf = in->private_data;
1216         unsigned int subbuf_size = rbuf->chan->subbuf_size;
1217         uint64_t pos = (uint64_t) *ppos;
1218         uint32_t alloc_size = (uint32_t) rbuf->chan->alloc_size;
1219         size_t read_start = (size_t) do_div(pos, alloc_size);
1220         size_t read_subbuf = read_start / subbuf_size;
1221         size_t padding = rbuf->padding[read_subbuf];
1222         size_t nonpad_end = read_subbuf * subbuf_size + subbuf_size - padding;
1223         struct page *pages[PIPE_DEF_BUFFERS];
1224         struct partial_page partial[PIPE_DEF_BUFFERS];
1225         struct splice_pipe_desc spd = {
1226                 .pages = pages,
1227                 .nr_pages = 0,
1228                 .nr_pages_max = PIPE_DEF_BUFFERS,
1229                 .partial = partial,
1230                 .flags = flags,
1231                 .ops = &relay_pipe_buf_ops,
1232                 .spd_release = relay_page_release,
1233         };
1234         ssize_t ret;
1235
1236         if (rbuf->subbufs_produced == rbuf->subbufs_consumed)
1237                 return 0;
1238         if (splice_grow_spd(pipe, &spd))
1239                 return -ENOMEM;
1240
1241         /*
1242          * Adjust read len, if longer than what is available
1243          */
1244         if (len > (subbuf_size - read_start % subbuf_size))
1245                 len = subbuf_size - read_start % subbuf_size;
1246
1247         subbuf_pages = rbuf->chan->alloc_size >> PAGE_SHIFT;
1248         pidx = (read_start / PAGE_SIZE) % subbuf_pages;
1249         poff = read_start & ~PAGE_MASK;
1250         nr_pages = min_t(unsigned int, subbuf_pages, spd.nr_pages_max);
1251
1252         for (total_len = 0; spd.nr_pages < nr_pages; spd.nr_pages++) {
1253                 unsigned int this_len, this_end, private;
1254                 unsigned int cur_pos = read_start + total_len;
1255
1256                 if (!len)
1257                         break;
1258
1259                 this_len = min_t(unsigned long, len, PAGE_SIZE - poff);
1260                 private = this_len;
1261
1262                 spd.pages[spd.nr_pages] = rbuf->page_array[pidx];
1263                 spd.partial[spd.nr_pages].offset = poff;
1264
1265                 this_end = cur_pos + this_len;
1266                 if (this_end >= nonpad_end) {
1267                         this_len = nonpad_end - cur_pos;
1268                         private = this_len + padding;
1269                 }
1270                 spd.partial[spd.nr_pages].len = this_len;
1271                 spd.partial[spd.nr_pages].private = private;
1272
1273                 len -= this_len;
1274                 total_len += this_len;
1275                 poff = 0;
1276                 pidx = (pidx + 1) % subbuf_pages;
1277
1278                 if (this_end >= nonpad_end) {
1279                         spd.nr_pages++;
1280                         break;
1281                 }
1282         }
1283
1284         ret = 0;
1285         if (!spd.nr_pages)
1286                 goto out;
1287
1288         ret = *nonpad_ret = splice_to_pipe(pipe, &spd);
1289         if (ret < 0 || ret < total_len)
1290                 goto out;
1291
1292         if (read_start + ret == nonpad_end)
1293                 ret += padding;
1294
1295 out:
1296         splice_shrink_spd(&spd);
1297         return ret;
1298 }
1299
1300 static ssize_t relay_file_splice_read(struct file *in,
1301                                       loff_t *ppos,
1302                                       struct pipe_inode_info *pipe,
1303                                       size_t len,
1304                                       unsigned int flags)
1305 {
1306         ssize_t spliced;
1307         int ret;
1308         int nonpad_ret = 0;
1309
1310         ret = 0;
1311         spliced = 0;
1312
1313         while (len && !spliced) {
1314                 ret = subbuf_splice_actor(in, ppos, pipe, len, flags, &nonpad_ret);
1315                 if (ret < 0)
1316                         break;
1317                 else if (!ret) {
1318                         if (flags & SPLICE_F_NONBLOCK)
1319                                 ret = -EAGAIN;
1320                         break;
1321                 }
1322
1323                 *ppos += ret;
1324                 if (ret > len)
1325                         len = 0;
1326                 else
1327                         len -= ret;
1328                 spliced += nonpad_ret;
1329                 nonpad_ret = 0;
1330         }
1331
1332         if (spliced)
1333                 return spliced;
1334
1335         return ret;
1336 }
1337
1338 const struct file_operations relay_file_operations = {
1339         .open           = relay_file_open,
1340         .poll           = relay_file_poll,
1341         .mmap           = relay_file_mmap,
1342         .read           = relay_file_read,
1343         .llseek         = no_llseek,
1344         .release        = relay_file_release,
1345         .splice_read    = relay_file_splice_read,
1346 };
1347 EXPORT_SYMBOL_GPL(relay_file_operations);
1348
1349 static __init int relay_init(void)
1350 {
1351
1352         hotcpu_notifier(relay_hotcpu_callback, 0);
1353         return 0;
1354 }
1355
1356 early_initcall(relay_init);