kernel/relay.c

   1 /*
   2  * Public API and common code for kernel->userspace relay file support.
   3  *
   4  * See Documentation/filesystems/relay.rst 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  * fault() vm_op implementation for relay file mapping.
  32  */
  33 static vm_fault_t relay_buf_fault(struct vm_fault *vmf)
  34 {
  35         struct page *page;
  36         struct rchan_buf *buf = vmf->vma->vm_private_data;
  37         pgoff_t pgoff = vmf->pgoff;
  38
  39         if (!buf)
  40                 return VM_FAULT_OOM;
  41
  42         page = vmalloc_to_page(buf->start + (pgoff << PAGE_SHIFT));
  43         if (!page)
  44                 return VM_FAULT_SIGBUS;
  45         get_page(page);
  46         vmf->page = page;
  47
  48         return 0;
  49 }
  50
  51 /*
  52  * vm_ops for relay file mappings.
  53  */
  54 static const struct vm_operations_struct relay_file_mmap_ops = {
  55         .fault = relay_buf_fault,
  56 };
  57
  58 /*
  59  * allocate an array of pointers of struct page
  60  */
  61 static struct page **relay_alloc_page_array(unsigned int n_pages)
  62 {
  63         const size_t pa_size = n_pages * sizeof(struct page *);
  64         if (pa_size > PAGE_SIZE)
  65                 return vzalloc(pa_size);
  66         return kzalloc(pa_size, GFP_KERNEL);
  67 }
  68
  69 /*
  70  * free an array of pointers of struct page
  71  */
  72 static void relay_free_page_array(struct page **array)
  73 {
  74         kvfree(array);
  75 }
  76
  77 /**
  78  *      relay_mmap_buf: - mmap channel buffer to process address space
  79  *      @buf: relay channel buffer
  80  *      @vma: vm_area_struct describing memory to be mapped
  81  *
  82  *      Returns 0 if ok, negative on error
  83  *
  84  *      Caller should already have grabbed mmap_lock.
  85  */
  86 static int relay_mmap_buf(struct rchan_buf *buf, struct vm_area_struct *vma)
  87 {
  88         unsigned long length = vma->vm_end - vma->vm_start;
  89
  90         if (!buf)
  91                 return -EBADF;
  92
  93         if (length != (unsigned long)buf->chan->alloc_size)
  94                 return -EINVAL;
  95
  96         vma->vm_ops = &relay_file_mmap_ops;
  97         vma->vm_flags |= VM_DONTEXPAND;
  98         vma->vm_private_data = buf;
  99
 100         return 0;
 101 }
 102
 103 /**
 104  *      relay_alloc_buf - allocate a channel buffer
 105  *      @buf: the buffer struct
 106  *      @size: total size of the buffer
 107  *
 108  *      Returns a pointer to the resulting buffer, %NULL if unsuccessful. The
 109  *      passed in size will get page aligned, if it isn't already.
 110  */
 111 static void *relay_alloc_buf(struct rchan_buf *buf, size_t *size)
 112 {
 113         void *mem;
 114         unsigned int i, j, n_pages;
 115
 116         *size = PAGE_ALIGN(*size);
 117         n_pages = *size >> PAGE_SHIFT;
 118
 119         buf->page_array = relay_alloc_page_array(n_pages);
 120         if (!buf->page_array)
 121                 return NULL;
 122
 123         for (i = 0; i < n_pages; i++) {
 124                 buf->page_array[i] = alloc_page(GFP_KERNEL);
 125                 if (unlikely(!buf->page_array[i]))
 126                         goto depopulate;
 127                 set_page_private(buf->page_array[i], (unsigned long)buf);
 128         }
 129         mem = vmap(buf->page_array, n_pages, VM_MAP, PAGE_KERNEL);
 130         if (!mem)
 131                 goto depopulate;
 132
 133         memset(mem, 0, *size);
 134         buf->page_count = n_pages;
 135         return mem;
 136
 137 depopulate:
 138         for (j = 0; j < i; j++)
 139                 __free_page(buf->page_array[j]);
 140         relay_free_page_array(buf->page_array);
 141         return NULL;
 142 }
 143
 144 /**
 145  *      relay_create_buf - allocate and initialize a channel buffer
 146  *      @chan: the relay channel
 147  *
 148  *      Returns channel buffer if successful, %NULL otherwise.
 149  */
 150 static struct rchan_buf *relay_create_buf(struct rchan *chan)
 151 {
 152         struct rchan_buf *buf;
 153
 154         if (chan->n_subbufs > KMALLOC_MAX_SIZE / sizeof(size_t *))
 155                 return NULL;
 156
 157         buf = kzalloc(sizeof(struct rchan_buf), GFP_KERNEL);
 158         if (!buf)
 159                 return NULL;
 160         buf->padding = kmalloc_array(chan->n_subbufs, sizeof(size_t *),
 161                                      GFP_KERNEL);
 162         if (!buf->padding)
 163                 goto free_buf;
 164
 165         buf->start = relay_alloc_buf(buf, &chan->alloc_size);
 166         if (!buf->start)
 167                 goto free_buf;
 168
 169         buf->chan = chan;
 170         kref_get(&buf->chan->kref);
 171         return buf;
 172
 173 free_buf:
 174         kfree(buf->padding);
 175         kfree(buf);
 176         return NULL;
 177 }
 178
 179 /**
 180  *      relay_destroy_channel - free the channel struct
 181  *      @kref: target kernel reference that contains the relay channel
 182  *
 183  *      Should only be called from kref_put().
 184  */
 185 static void relay_destroy_channel(struct kref *kref)
 186 {
 187         struct rchan *chan = container_of(kref, struct rchan, kref);
 188         free_percpu(chan->buf);
 189         kfree(chan);
 190 }
 191
 192 /**
 193  *      relay_destroy_buf - destroy an rchan_buf struct and associated buffer
 194  *      @buf: the buffer struct
 195  */
 196 static void relay_destroy_buf(struct rchan_buf *buf)
 197 {
 198         struct rchan *chan = buf->chan;
 199         unsigned int i;
 200
 201         if (likely(buf->start)) {
 202                 vunmap(buf->start);
 203                 for (i = 0; i < buf->page_count; i++)
 204                         __free_page(buf->page_array[i]);
 205                 relay_free_page_array(buf->page_array);
 206         }
 207         *per_cpu_ptr(chan->buf, buf->cpu) = NULL;
 208         kfree(buf->padding);
 209         kfree(buf);
 210         kref_put(&chan->kref, relay_destroy_channel);
 211 }
 212
 213 /**
 214  *      relay_remove_buf - remove a channel buffer
 215  *      @kref: target kernel reference that contains the relay buffer
 216  *
 217  *      Removes the file from the filesystem, which also frees the
 218  *      rchan_buf_struct and the channel buffer.  Should only be called from
 219  *      kref_put().
 220  */
 221 static void relay_remove_buf(struct kref *kref)
 222 {
 223         struct rchan_buf *buf = container_of(kref, struct rchan_buf, kref);
 224         relay_destroy_buf(buf);
 225 }
 226
 227 /**
 228  *      relay_buf_empty - boolean, is the channel buffer empty?
 229  *      @buf: channel buffer
 230  *
 231  *      Returns 1 if the buffer is empty, 0 otherwise.
 232  */
 233 static int relay_buf_empty(struct rchan_buf *buf)
 234 {
 235         return (buf->subbufs_produced - buf->subbufs_consumed) ? 0 : 1;
 236 }
 237
 238 /**
 239  *      relay_buf_full - boolean, is the channel buffer full?
 240  *      @buf: channel buffer
 241  *
 242  *      Returns 1 if the buffer is full, 0 otherwise.
 243  */
 244 int relay_buf_full(struct rchan_buf *buf)
 245 {
 246         size_t ready = buf->subbufs_produced - buf->subbufs_consumed;
 247         return (ready >= buf->chan->n_subbufs) ? 1 : 0;
 248 }
 249 EXPORT_SYMBOL_GPL(relay_buf_full);
 250
 251 /*
 252  * High-level relay kernel API and associated functions.
 253  */
 254
 255 static int relay_subbuf_start(struct rchan_buf *buf, void *subbuf,
 256                               void *prev_subbuf, size_t prev_padding)
 257 {
 258         if (!buf->chan->cb->subbuf_start)
 259                 return !relay_buf_full(buf);
 260
 261         return buf->chan->cb->subbuf_start(buf, subbuf,
 262                                            prev_subbuf, prev_padding);
 263 }
 264
 265 /**
 266  *      wakeup_readers - wake up readers waiting on a channel
 267  *      @work: contains the channel buffer
 268  *
 269  *      This is the function used to defer reader waking
 270  */
 271 static void wakeup_readers(struct irq_work *work)
 272 {
 273         struct rchan_buf *buf;
 274
 275         buf = container_of(work, struct rchan_buf, wakeup_work);
 276         wake_up_interruptible(&buf->read_wait);
 277 }
 278
 279 /**
 280  *      __relay_reset - reset a channel buffer
 281  *      @buf: the channel buffer
 282  *      @init: 1 if this is a first-time initialization
 283  *
 284  *      See relay_reset() for description of effect.
 285  */
 286 static void __relay_reset(struct rchan_buf *buf, unsigned int init)
 287 {
 288         size_t i;
 289
 290         if (init) {
 291                 init_waitqueue_head(&buf->read_wait);
 292                 kref_init(&buf->kref);
 293                 init_irq_work(&buf->wakeup_work, wakeup_readers);
 294         } else {
 295                 irq_work_sync(&buf->wakeup_work);
 296         }
 297
 298         buf->subbufs_produced = 0;
 299         buf->subbufs_consumed = 0;
 300         buf->bytes_consumed = 0;
 301         buf->finalized = 0;
 302         buf->data = buf->start;
 303         buf->offset = 0;
 304
 305         for (i = 0; i < buf->chan->n_subbufs; i++)
 306                 buf->padding[i] = 0;
 307
 308         relay_subbuf_start(buf, buf->data, NULL, 0);
 309 }
 310
 311 /**
 312  *      relay_reset - reset the channel
 313  *      @chan: the channel
 314  *
 315  *      This has the effect of erasing all data from all channel buffers
 316  *      and restarting the channel in its initial state.  The buffers
 317  *      are not freed, so any mappings are still in effect.
 318  *
 319  *      NOTE. Care should be taken that the channel isn't actually
 320  *      being used by anything when this call is made.
 321  */
 322 void relay_reset(struct rchan *chan)
 323 {
 324         struct rchan_buf *buf;
 325         unsigned int i;
 326
 327         if (!chan)
 328                 return;
 329
 330         if (chan->is_global && (buf = *per_cpu_ptr(chan->buf, 0))) {
 331                 __relay_reset(buf, 0);
 332                 return;
 333         }
 334
 335         mutex_lock(&relay_channels_mutex);
 336         for_each_possible_cpu(i)
 337                 if ((buf = *per_cpu_ptr(chan->buf, i)))
 338                         __relay_reset(buf, 0);
 339         mutex_unlock(&relay_channels_mutex);
 340 }
 341 EXPORT_SYMBOL_GPL(relay_reset);
 342
 343 static inline void relay_set_buf_dentry(struct rchan_buf *buf,
 344                                         struct dentry *dentry)
 345 {
 346         buf->dentry = dentry;
 347         d_inode(buf->dentry)->i_size = buf->early_bytes;
 348 }
 349
 350 static struct dentry *relay_create_buf_file(struct rchan *chan,
 351                                             struct rchan_buf *buf,
 352                                             unsigned int cpu)
 353 {
 354         struct dentry *dentry;
 355         char *tmpname;
 356
 357         tmpname = kzalloc(NAME_MAX + 1, GFP_KERNEL);
 358         if (!tmpname)
 359                 return NULL;
 360         snprintf(tmpname, NAME_MAX, "%s%d", chan->base_filename, cpu);
 361
 362         /* Create file in fs */
 363         dentry = chan->cb->create_buf_file(tmpname, chan->parent,
 364                                            S_IRUSR, buf,
 365                                            &chan->is_global);
 366         if (IS_ERR(dentry))
 367                 dentry = NULL;
 368
 369         kfree(tmpname);
 370
 371         return dentry;
 372 }
 373
 374 /*
 375  *      relay_open_buf - create a new relay channel buffer
 376  *
 377  *      used by relay_open() and CPU hotplug.
 378  */
 379 static struct rchan_buf *relay_open_buf(struct rchan *chan, unsigned int cpu)
 380 {
 381         struct rchan_buf *buf = NULL;
 382         struct dentry *dentry;
 383
 384         if (chan->is_global)
 385                 return *per_cpu_ptr(chan->buf, 0);
 386
 387         buf = relay_create_buf(chan);
 388         if (!buf)
 389                 return NULL;
 390
 391         if (chan->has_base_filename) {
 392                 dentry = relay_create_buf_file(chan, buf, cpu);
 393                 if (!dentry)
 394                         goto free_buf;
 395                 relay_set_buf_dentry(buf, dentry);
 396         } else {
 397                 /* Only retrieve global info, nothing more, nothing less */
 398                 dentry = chan->cb->create_buf_file(NULL, NULL,
 399                                                    S_IRUSR, buf,
 400                                                    &chan->is_global);
 401                 if (IS_ERR_OR_NULL(dentry))
 402                         goto free_buf;
 403         }
 404
 405         buf->cpu = cpu;
 406         __relay_reset(buf, 1);
 407
 408         if(chan->is_global) {
 409                 *per_cpu_ptr(chan->buf, 0) = buf;
 410                 buf->cpu = 0;
 411         }
 412
 413         return buf;
 414
 415 free_buf:
 416         relay_destroy_buf(buf);
 417         return NULL;
 418 }
 419
 420 /**
 421  *      relay_close_buf - close a channel buffer
 422  *      @buf: channel buffer
 423  *
 424  *      Marks the buffer finalized and restores the default callbacks.
 425  *      The channel buffer and channel buffer data structure are then freed
 426  *      automatically when the last reference is given up.
 427  */
 428 static void relay_close_buf(struct rchan_buf *buf)
 429 {
 430         buf->finalized = 1;
 431         irq_work_sync(&buf->wakeup_work);
 432         buf->chan->cb->remove_buf_file(buf->dentry);
 433         kref_put(&buf->kref, relay_remove_buf);
 434 }
 435
 436 int relay_prepare_cpu(unsigned int cpu)
 437 {
 438         struct rchan *chan;
 439         struct rchan_buf *buf;
 440
 441         mutex_lock(&relay_channels_mutex);
 442         list_for_each_entry(chan, &relay_channels, list) {
 443                 if (*per_cpu_ptr(chan->buf, cpu))
 444                         continue;
 445                 buf = relay_open_buf(chan, cpu);
 446                 if (!buf) {
 447                         pr_err("relay: cpu %d buffer creation failed\n", cpu);
 448                         mutex_unlock(&relay_channels_mutex);
 449                         return -ENOMEM;
 450                 }
 451                 *per_cpu_ptr(chan->buf, cpu) = buf;
 452         }
 453         mutex_unlock(&relay_channels_mutex);
 454         return 0;
 455 }
 456
 457 /**
 458  *      relay_open - create a new relay channel
 459  *      @base_filename: base name of files to create, %NULL for buffering only
 460  *      @parent: dentry of parent directory, %NULL for root directory or buffer
 461  *      @subbuf_size: size of sub-buffers
 462  *      @n_subbufs: number of sub-buffers
 463  *      @cb: client callback functions
 464  *      @private_data: user-defined data
 465  *
 466  *      Returns channel pointer if successful, %NULL otherwise.
 467  *
 468  *      Creates a channel buffer for each cpu using the sizes and
 469  *      attributes specified.  The created channel buffer files
 470  *      will be named base_filename0...base_filenameN-1.  File
 471  *      permissions will be %S_IRUSR.
 472  *
 473  *      If opening a buffer (@parent = NULL) that you later wish to register
 474  *      in a filesystem, call relay_late_setup_files() once the @parent dentry
 475  *      is available.
 476  */
 477 struct rchan *relay_open(const char *base_filename,
 478                          struct dentry *parent,
 479                          size_t subbuf_size,
 480                          size_t n_subbufs,
 481                          const struct rchan_callbacks *cb,
 482                          void *private_data)
 483 {
 484         unsigned int i;
 485         struct rchan *chan;
 486         struct rchan_buf *buf;
 487
 488         if (!(subbuf_size && n_subbufs))
 489                 return NULL;
 490         if (subbuf_size > UINT_MAX / n_subbufs)
 491                 return NULL;
 492         if (!cb || !cb->create_buf_file || !cb->remove_buf_file)
 493                 return NULL;
 494
 495         chan = kzalloc(sizeof(struct rchan), GFP_KERNEL);
 496         if (!chan)
 497                 return NULL;
 498
 499         chan->buf = alloc_percpu(struct rchan_buf *);
 500         if (!chan->buf) {
 501                 kfree(chan);
 502                 return NULL;
 503         }
 504
 505         chan->version = RELAYFS_CHANNEL_VERSION;
 506         chan->n_subbufs = n_subbufs;
 507         chan->subbuf_size = subbuf_size;
 508         chan->alloc_size = PAGE_ALIGN(subbuf_size * n_subbufs);
 509         chan->parent = parent;
 510         chan->private_data = private_data;
 511         if (base_filename) {
 512                 chan->has_base_filename = 1;
 513                 strlcpy(chan->base_filename, base_filename, NAME_MAX);
 514         }
 515         chan->cb = cb;
 516         kref_init(&chan->kref);
 517
 518         mutex_lock(&relay_channels_mutex);
 519         for_each_online_cpu(i) {
 520                 buf = relay_open_buf(chan, i);
 521                 if (!buf)
 522                         goto free_bufs;
 523                 *per_cpu_ptr(chan->buf, i) = buf;
 524         }
 525         list_add(&chan->list, &relay_channels);
 526         mutex_unlock(&relay_channels_mutex);
 527
 528         return chan;
 529
 530 free_bufs:
 531         for_each_possible_cpu(i) {
 532                 if ((buf = *per_cpu_ptr(chan->buf, i)))
 533                         relay_close_buf(buf);
 534         }
 535
 536         kref_put(&chan->kref, relay_destroy_channel);
 537         mutex_unlock(&relay_channels_mutex);
 538         return NULL;
 539 }
 540 EXPORT_SYMBOL_GPL(relay_open);
 541
 542 struct rchan_percpu_buf_dispatcher {
 543         struct rchan_buf *buf;
 544         struct dentry *dentry;
 545 };
 546
 547 /* Called in atomic context. */
 548 static void __relay_set_buf_dentry(void *info)
 549 {
 550         struct rchan_percpu_buf_dispatcher *p = info;
 551
 552         relay_set_buf_dentry(p->buf, p->dentry);
 553 }
 554
 555 /**
 556  *      relay_late_setup_files - triggers file creation
 557  *      @chan: channel to operate on
 558  *      @base_filename: base name of files to create
 559  *      @parent: dentry of parent directory, %NULL for root directory
 560  *
 561  *      Returns 0 if successful, non-zero otherwise.
 562  *
 563  *      Use to setup files for a previously buffer-only channel created
 564  *      by relay_open() with a NULL parent dentry.
 565  *
 566  *      For example, this is useful for perfomring early tracing in kernel,
 567  *      before VFS is up and then exposing the early results once the dentry
 568  *      is available.
 569  */
 570 int relay_late_setup_files(struct rchan *chan,
 571                            const char *base_filename,
 572                            struct dentry *parent)
 573 {
 574         int err = 0;
 575         unsigned int i, curr_cpu;
 576         unsigned long flags;
 577         struct dentry *dentry;
 578         struct rchan_buf *buf;
 579         struct rchan_percpu_buf_dispatcher disp;
 580
 581         if (!chan || !base_filename)
 582                 return -EINVAL;
 583
 584         strlcpy(chan->base_filename, base_filename, NAME_MAX);
 585
 586         mutex_lock(&relay_channels_mutex);
 587         /* Is chan already set up? */
 588         if (unlikely(chan->has_base_filename)) {
 589                 mutex_unlock(&relay_channels_mutex);
 590                 return -EEXIST;
 591         }
 592         chan->has_base_filename = 1;
 593         chan->parent = parent;
 594
 595         if (chan->is_global) {
 596                 err = -EINVAL;
 597                 buf = *per_cpu_ptr(chan->buf, 0);
 598                 if (!WARN_ON_ONCE(!buf)) {
 599                         dentry = relay_create_buf_file(chan, buf, 0);
 600                         if (dentry && !WARN_ON_ONCE(!chan->is_global)) {
 601                                 relay_set_buf_dentry(buf, dentry);
 602                                 err = 0;
 603                         }
 604                 }
 605                 mutex_unlock(&relay_channels_mutex);
 606                 return err;
 607         }
 608
 609         curr_cpu = get_cpu();
 610         /*
 611          * The CPU hotplug notifier ran before us and created buffers with
 612          * no files associated. So it's safe to call relay_setup_buf_file()
 613          * on all currently online CPUs.
 614          */
 615         for_each_online_cpu(i) {
 616                 buf = *per_cpu_ptr(chan->buf, i);
 617                 if (unlikely(!buf)) {
 618                         WARN_ONCE(1, KERN_ERR "CPU has no buffer!\n");
 619                         err = -EINVAL;
 620                         break;
 621                 }
 622
 623                 dentry = relay_create_buf_file(chan, buf, i);
 624                 if (unlikely(!dentry)) {
 625                         err = -EINVAL;
 626                         break;
 627                 }
 628
 629                 if (curr_cpu == i) {
 630                         local_irq_save(flags);
 631                         relay_set_buf_dentry(buf, dentry);
 632                         local_irq_restore(flags);
 633                 } else {
 634                         disp.buf = buf;
 635                         disp.dentry = dentry;
 636                         smp_mb();
 637                         /* relay_channels_mutex must be held, so wait. */
 638                         err = smp_call_function_single(i,
 639                                                        __relay_set_buf_dentry,
 640                                                        &disp, 1);
 641                 }
 642                 if (unlikely(err))
 643                         break;
 644         }
 645         put_cpu();
 646         mutex_unlock(&relay_channels_mutex);
 647
 648         return err;
 649 }
 650 EXPORT_SYMBOL_GPL(relay_late_setup_files);
 651
 652 /**
 653  *      relay_switch_subbuf - switch to a new sub-buffer
 654  *      @buf: channel buffer
 655  *      @length: size of current event
 656  *
 657  *      Returns either the length passed in or 0 if full.
 658  *
 659  *      Performs sub-buffer-switch tasks such as invoking callbacks,
 660  *      updating padding counts, waking up readers, etc.
 661  */
 662 size_t relay_switch_subbuf(struct rchan_buf *buf, size_t length)
 663 {
 664         void *old, *new;
 665         size_t old_subbuf, new_subbuf;
 666
 667         if (unlikely(length > buf->chan->subbuf_size))
 668                 goto toobig;
 669
 670         if (buf->offset != buf->chan->subbuf_size + 1) {
 671                 buf->prev_padding = buf->chan->subbuf_size - buf->offset;
 672                 old_subbuf = buf->subbufs_produced % buf->chan->n_subbufs;
 673                 buf->padding[old_subbuf] = buf->prev_padding;
 674                 buf->subbufs_produced++;
 675                 if (buf->dentry)
 676                         d_inode(buf->dentry)->i_size +=
 677                                 buf->chan->subbuf_size -
 678                                 buf->padding[old_subbuf];
 679                 else
 680                         buf->early_bytes += buf->chan->subbuf_size -
 681                                             buf->padding[old_subbuf];
 682                 smp_mb();
 683                 if (waitqueue_active(&buf->read_wait)) {
 684                         /*
 685                          * Calling wake_up_interruptible() from here
 686                          * will deadlock if we happen to be logging
 687                          * from the scheduler (trying to re-grab
 688                          * rq->lock), so defer it.
 689                          */
 690                         irq_work_queue(&buf->wakeup_work);
 691                 }
 692         }
 693
 694         old = buf->data;
 695         new_subbuf = buf->subbufs_produced % buf->chan->n_subbufs;
 696         new = buf->start + new_subbuf * buf->chan->subbuf_size;
 697         buf->offset = 0;
 698         if (!relay_subbuf_start(buf, new, old, buf->prev_padding)) {
 699                 buf->offset = buf->chan->subbuf_size + 1;
 700                 return 0;
 701         }
 702         buf->data = new;
 703         buf->padding[new_subbuf] = 0;
 704
 705         if (unlikely(length + buf->offset > buf->chan->subbuf_size))
 706                 goto toobig;
 707
 708         return length;
 709
 710 toobig:
 711         buf->chan->last_toobig = length;
 712         return 0;
 713 }
 714 EXPORT_SYMBOL_GPL(relay_switch_subbuf);
 715
 716 /**
 717  *      relay_subbufs_consumed - update the buffer's sub-buffers-consumed count
 718  *      @chan: the channel
 719  *      @cpu: the cpu associated with the channel buffer to update
 720  *      @subbufs_consumed: number of sub-buffers to add to current buf's count
 721  *
 722  *      Adds to the channel buffer's consumed sub-buffer count.
 723  *      subbufs_consumed should be the number of sub-buffers newly consumed,
 724  *      not the total consumed.
 725  *
 726  *      NOTE. Kernel clients don't need to call this function if the channel
 727  *      mode is 'overwrite'.
 728  */
 729 void relay_subbufs_consumed(struct rchan *chan,
 730                             unsigned int cpu,
 731                             size_t subbufs_consumed)
 732 {
 733         struct rchan_buf *buf;
 734
 735         if (!chan || cpu >= NR_CPUS)
 736                 return;
 737
 738         buf = *per_cpu_ptr(chan->buf, cpu);
 739         if (!buf || subbufs_consumed > chan->n_subbufs)
 740                 return;
 741
 742         if (subbufs_consumed > buf->subbufs_produced - buf->subbufs_consumed)
 743                 buf->subbufs_consumed = buf->subbufs_produced;
 744         else
 745                 buf->subbufs_consumed += subbufs_consumed;
 746 }
 747 EXPORT_SYMBOL_GPL(relay_subbufs_consumed);
 748
 749 /**
 750  *      relay_close - close the channel
 751  *      @chan: the channel
 752  *
 753  *      Closes all channel buffers and frees the channel.
 754  */
 755 void relay_close(struct rchan *chan)
 756 {
 757         struct rchan_buf *buf;
 758         unsigned int i;
 759
 760         if (!chan)
 761                 return;
 762
 763         mutex_lock(&relay_channels_mutex);
 764         if (chan->is_global && (buf = *per_cpu_ptr(chan->buf, 0)))
 765                 relay_close_buf(buf);
 766         else
 767                 for_each_possible_cpu(i)
 768                         if ((buf = *per_cpu_ptr(chan->buf, i)))
 769                                 relay_close_buf(buf);
 770
 771         if (chan->last_toobig)
 772                 printk(KERN_WARNING "relay: one or more items not logged "
 773                        "[item size (%zd) > sub-buffer size (%zd)]\n",
 774                        chan->last_toobig, chan->subbuf_size);
 775
 776         list_del(&chan->list);
 777         kref_put(&chan->kref, relay_destroy_channel);
 778         mutex_unlock(&relay_channels_mutex);
 779 }
 780 EXPORT_SYMBOL_GPL(relay_close);
 781
 782 /**
 783  *      relay_flush - close the channel
 784  *      @chan: the channel
 785  *
 786  *      Flushes all channel buffers, i.e. forces buffer switch.
 787  */
 788 void relay_flush(struct rchan *chan)
 789 {
 790         struct rchan_buf *buf;
 791         unsigned int i;
 792
 793         if (!chan)
 794                 return;
 795
 796         if (chan->is_global && (buf = *per_cpu_ptr(chan->buf, 0))) {
 797                 relay_switch_subbuf(buf, 0);
 798                 return;
 799         }
 800
 801         mutex_lock(&relay_channels_mutex);
 802         for_each_possible_cpu(i)
 803                 if ((buf = *per_cpu_ptr(chan->buf, i)))
 804                         relay_switch_subbuf(buf, 0);
 805         mutex_unlock(&relay_channels_mutex);
 806 }
 807 EXPORT_SYMBOL_GPL(relay_flush);
 808
 809 /**
 810  *      relay_file_open - open file op for relay files
 811  *      @inode: the inode
 812  *      @filp: the file
 813  *
 814  *      Increments the channel buffer refcount.
 815  */
 816 static int relay_file_open(struct inode *inode, struct file *filp)
 817 {
 818         struct rchan_buf *buf = inode->i_private;
 819         kref_get(&buf->kref);
 820         filp->private_data = buf;
 821
 822         return nonseekable_open(inode, filp);
 823 }
 824
 825 /**
 826  *      relay_file_mmap - mmap file op for relay files
 827  *      @filp: the file
 828  *      @vma: the vma describing what to map
 829  *
 830  *      Calls upon relay_mmap_buf() to map the file into user space.
 831  */
 832 static int relay_file_mmap(struct file *filp, struct vm_area_struct *vma)
 833 {
 834         struct rchan_buf *buf = filp->private_data;
 835         return relay_mmap_buf(buf, vma);
 836 }
 837
 838 /**
 839  *      relay_file_poll - poll file op for relay files
 840  *      @filp: the file
 841  *      @wait: poll table
 842  *
 843  *      Poll implemention.
 844  */
 845 static __poll_t relay_file_poll(struct file *filp, poll_table *wait)
 846 {
 847         __poll_t mask = 0;
 848         struct rchan_buf *buf = filp->private_data;
 849
 850         if (buf->finalized)
 851                 return EPOLLERR;
 852
 853         if (filp->f_mode & FMODE_READ) {
 854                 poll_wait(filp, &buf->read_wait, wait);
 855                 if (!relay_buf_empty(buf))
 856                         mask |= EPOLLIN | EPOLLRDNORM;
 857         }
 858
 859         return mask;
 860 }
 861
 862 /**
 863  *      relay_file_release - release file op for relay files
 864  *      @inode: the inode
 865  *      @filp: the file
 866  *
 867  *      Decrements the channel refcount, as the filesystem is
 868  *      no longer using it.
 869  */
 870 static int relay_file_release(struct inode *inode, struct file *filp)
 871 {
 872         struct rchan_buf *buf = filp->private_data;
 873         kref_put(&buf->kref, relay_remove_buf);
 874
 875         return 0;
 876 }
 877
 878 /*
 879  *      relay_file_read_consume - update the consumed count for the buffer
 880  */
 881 static void relay_file_read_consume(struct rchan_buf *buf,
 882                                     size_t read_pos,
 883                                     size_t bytes_consumed)
 884 {
 885         size_t subbuf_size = buf->chan->subbuf_size;
 886         size_t n_subbufs = buf->chan->n_subbufs;
 887         size_t read_subbuf;
 888
 889         if (buf->subbufs_produced == buf->subbufs_consumed &&
 890             buf->offset == buf->bytes_consumed)
 891                 return;
 892
 893         if (buf->bytes_consumed + bytes_consumed > subbuf_size) {
 894                 relay_subbufs_consumed(buf->chan, buf->cpu, 1);
 895                 buf->bytes_consumed = 0;
 896         }
 897
 898         buf->bytes_consumed += bytes_consumed;
 899         if (!read_pos)
 900                 read_subbuf = buf->subbufs_consumed % n_subbufs;
 901         else
 902                 read_subbuf = read_pos / buf->chan->subbuf_size;
 903         if (buf->bytes_consumed + buf->padding[read_subbuf] == subbuf_size) {
 904                 if ((read_subbuf == buf->subbufs_produced % n_subbufs) &&
 905                     (buf->offset == subbuf_size))
 906                         return;
 907                 relay_subbufs_consumed(buf->chan, buf->cpu, 1);
 908                 buf->bytes_consumed = 0;
 909         }
 910 }
 911
 912 /*
 913  *      relay_file_read_avail - boolean, are there unconsumed bytes available?
 914  */
 915 static int relay_file_read_avail(struct rchan_buf *buf)
 916 {
 917         size_t subbuf_size = buf->chan->subbuf_size;
 918         size_t n_subbufs = buf->chan->n_subbufs;
 919         size_t produced = buf->subbufs_produced;
 920         size_t consumed;
 921
 922         relay_file_read_consume(buf, 0, 0);
 923
 924         consumed = buf->subbufs_consumed;
 925
 926         if (unlikely(buf->offset > subbuf_size)) {
 927                 if (produced == consumed)
 928                         return 0;
 929                 return 1;
 930         }
 931
 932         if (unlikely(produced - consumed >= n_subbufs)) {
 933                 consumed = produced - n_subbufs + 1;
 934                 buf->subbufs_consumed = consumed;
 935                 buf->bytes_consumed = 0;
 936         }
 937
 938         produced = (produced % n_subbufs) * subbuf_size + buf->offset;
 939         consumed = (consumed % n_subbufs) * subbuf_size + buf->bytes_consumed;
 940
 941         if (consumed > produced)
 942                 produced += n_subbufs * subbuf_size;
 943
 944         if (consumed == produced) {
 945                 if (buf->offset == subbuf_size &&
 946                     buf->subbufs_produced > buf->subbufs_consumed)
 947                         return 1;
 948                 return 0;
 949         }
 950
 951         return 1;
 952 }
 953
 954 /**
 955  *      relay_file_read_subbuf_avail - return bytes available in sub-buffer
 956  *      @read_pos: file read position
 957  *      @buf: relay channel buffer
 958  */
 959 static size_t relay_file_read_subbuf_avail(size_t read_pos,
 960                                            struct rchan_buf *buf)
 961 {
 962         size_t padding, avail = 0;
 963         size_t read_subbuf, read_offset, write_subbuf, write_offset;
 964         size_t subbuf_size = buf->chan->subbuf_size;
 965
 966         write_subbuf = (buf->data - buf->start) / subbuf_size;
 967         write_offset = buf->offset > subbuf_size ? subbuf_size : buf->offset;
 968         read_subbuf = read_pos / subbuf_size;
 969         read_offset = read_pos % subbuf_size;
 970         padding = buf->padding[read_subbuf];
 971
 972         if (read_subbuf == write_subbuf) {
 973                 if (read_offset + padding < write_offset)
 974                         avail = write_offset - (read_offset + padding);
 975         } else
 976                 avail = (subbuf_size - padding) - read_offset;
 977
 978         return avail;
 979 }
 980
 981 /**
 982  *      relay_file_read_start_pos - find the first available byte to read
 983  *      @buf: relay channel buffer
 984  *
 985  *      If the read_pos is in the middle of padding, return the
 986  *      position of the first actually available byte, otherwise
 987  *      return the original value.
 988  */
 989 static size_t relay_file_read_start_pos(struct rchan_buf *buf)
 990 {
 991         size_t read_subbuf, padding, padding_start, padding_end;
 992         size_t subbuf_size = buf->chan->subbuf_size;
 993         size_t n_subbufs = buf->chan->n_subbufs;
 994         size_t consumed = buf->subbufs_consumed % n_subbufs;
 995         size_t read_pos = consumed * subbuf_size + buf->bytes_consumed;
 996
 997         read_subbuf = read_pos / subbuf_size;
 998         padding = buf->padding[read_subbuf];
 999         padding_start = (read_subbuf + 1) * subbuf_size - padding;
1000         padding_end = (read_subbuf + 1) * subbuf_size;
1001         if (read_pos >= padding_start && read_pos < padding_end) {
1002                 read_subbuf = (read_subbuf + 1) % n_subbufs;
1003                 read_pos = read_subbuf * subbuf_size;
1004         }
1005
1006         return read_pos;
1007 }
1008
1009 /**
1010  *      relay_file_read_end_pos - return the new read position
1011  *      @read_pos: file read position
1012  *      @buf: relay channel buffer
1013  *      @count: number of bytes to be read
1014  */
1015 static size_t relay_file_read_end_pos(struct rchan_buf *buf,
1016                                       size_t read_pos,
1017                                       size_t count)
1018 {
1019         size_t read_subbuf, padding, end_pos;
1020         size_t subbuf_size = buf->chan->subbuf_size;
1021         size_t n_subbufs = buf->chan->n_subbufs;
1022
1023         read_subbuf = read_pos / subbuf_size;
1024         padding = buf->padding[read_subbuf];
1025         if (read_pos % subbuf_size + count + padding == subbuf_size)
1026                 end_pos = (read_subbuf + 1) * subbuf_size;
1027         else
1028                 end_pos = read_pos + count;
1029         if (end_pos >= subbuf_size * n_subbufs)
1030                 end_pos = 0;
1031
1032         return end_pos;
1033 }
1034
1035 static ssize_t relay_file_read(struct file *filp,
1036                                char __user *buffer,
1037                                size_t count,
1038                                loff_t *ppos)
1039 {
1040         struct rchan_buf *buf = filp->private_data;
1041         size_t read_start, avail;
1042         size_t written = 0;
1043         int ret;
1044
1045         if (!count)
1046                 return 0;
1047
1048         inode_lock(file_inode(filp));
1049         do {
1050                 void *from;
1051
1052                 if (!relay_file_read_avail(buf))
1053                         break;
1054
1055                 read_start = relay_file_read_start_pos(buf);
1056                 avail = relay_file_read_subbuf_avail(read_start, buf);
1057                 if (!avail)
1058                         break;
1059
1060                 avail = min(count, avail);
1061                 from = buf->start + read_start;
1062                 ret = avail;
1063                 if (copy_to_user(buffer, from, avail))
1064                         break;
1065
1066                 buffer += ret;
1067                 written += ret;
1068                 count -= ret;
1069
1070                 relay_file_read_consume(buf, read_start, ret);
1071                 *ppos = relay_file_read_end_pos(buf, read_start, ret);
1072         } while (count);
1073         inode_unlock(file_inode(filp));
1074
1075         return written;
1076 }
1077
1078 static void relay_consume_bytes(struct rchan_buf *rbuf, int bytes_consumed)
1079 {
1080         rbuf->bytes_consumed += bytes_consumed;
1081
1082         if (rbuf->bytes_consumed >= rbuf->chan->subbuf_size) {
1083                 relay_subbufs_consumed(rbuf->chan, rbuf->cpu, 1);
1084                 rbuf->bytes_consumed %= rbuf->chan->subbuf_size;
1085         }
1086 }
1087
1088 static void relay_pipe_buf_release(struct pipe_inode_info *pipe,
1089                                    struct pipe_buffer *buf)
1090 {
1091         struct rchan_buf *rbuf;
1092
1093         rbuf = (struct rchan_buf *)page_private(buf->page);
1094         relay_consume_bytes(rbuf, buf->private);
1095 }
1096
1097 static const struct pipe_buf_operations relay_pipe_buf_ops = {
1098         .release        = relay_pipe_buf_release,
1099         .try_steal      = generic_pipe_buf_try_steal,
1100         .get            = generic_pipe_buf_get,
1101 };
1102
1103 static void relay_page_release(struct splice_pipe_desc *spd, unsigned int i)
1104 {
1105 }
1106
1107 /*
1108  *      subbuf_splice_actor - splice up to one subbuf's worth of data
1109  */
1110 static ssize_t subbuf_splice_actor(struct file *in,
1111                                loff_t *ppos,
1112                                struct pipe_inode_info *pipe,
1113                                size_t len,
1114                                unsigned int flags,
1115                                int *nonpad_ret)
1116 {
1117         unsigned int pidx, poff, total_len, subbuf_pages, nr_pages;
1118         struct rchan_buf *rbuf = in->private_data;
1119         unsigned int subbuf_size = rbuf->chan->subbuf_size;
1120         uint64_t pos = (uint64_t) *ppos;
1121         uint32_t alloc_size = (uint32_t) rbuf->chan->alloc_size;
1122         size_t read_start = (size_t) do_div(pos, alloc_size);
1123         size_t read_subbuf = read_start / subbuf_size;
1124         size_t padding = rbuf->padding[read_subbuf];
1125         size_t nonpad_end = read_subbuf * subbuf_size + subbuf_size - padding;
1126         struct page *pages[PIPE_DEF_BUFFERS];
1127         struct partial_page partial[PIPE_DEF_BUFFERS];
1128         struct splice_pipe_desc spd = {
1129                 .pages = pages,
1130                 .nr_pages = 0,
1131                 .nr_pages_max = PIPE_DEF_BUFFERS,
1132                 .partial = partial,
1133                 .ops = &relay_pipe_buf_ops,
1134                 .spd_release = relay_page_release,
1135         };
1136         ssize_t ret;
1137
1138         if (rbuf->subbufs_produced == rbuf->subbufs_consumed)
1139                 return 0;
1140         if (splice_grow_spd(pipe, &spd))
1141                 return -ENOMEM;
1142
1143         /*
1144          * Adjust read len, if longer than what is available
1145          */
1146         if (len > (subbuf_size - read_start % subbuf_size))
1147                 len = subbuf_size - read_start % subbuf_size;
1148
1149         subbuf_pages = rbuf->chan->alloc_size >> PAGE_SHIFT;
1150         pidx = (read_start / PAGE_SIZE) % subbuf_pages;
1151         poff = read_start & ~PAGE_MASK;
1152         nr_pages = min_t(unsigned int, subbuf_pages, spd.nr_pages_max);
1153
1154         for (total_len = 0; spd.nr_pages < nr_pages; spd.nr_pages++) {
1155                 unsigned int this_len, this_end, private;
1156                 unsigned int cur_pos = read_start + total_len;
1157
1158                 if (!len)
1159                         break;
1160
1161                 this_len = min_t(unsigned long, len, PAGE_SIZE - poff);
1162                 private = this_len;
1163
1164                 spd.pages[spd.nr_pages] = rbuf->page_array[pidx];
1165                 spd.partial[spd.nr_pages].offset = poff;
1166
1167                 this_end = cur_pos + this_len;
1168                 if (this_end >= nonpad_end) {
1169                         this_len = nonpad_end - cur_pos;
1170                         private = this_len + padding;
1171                 }
1172                 spd.partial[spd.nr_pages].len = this_len;
1173                 spd.partial[spd.nr_pages].private = private;
1174
1175                 len -= this_len;
1176                 total_len += this_len;
1177                 poff = 0;
1178                 pidx = (pidx + 1) % subbuf_pages;
1179
1180                 if (this_end >= nonpad_end) {
1181                         spd.nr_pages++;
1182                         break;
1183                 }
1184         }
1185
1186         ret = 0;
1187         if (!spd.nr_pages)
1188                 goto out;
1189
1190         ret = *nonpad_ret = splice_to_pipe(pipe, &spd);
1191         if (ret < 0 || ret < total_len)
1192                 goto out;
1193
1194         if (read_start + ret == nonpad_end)
1195                 ret += padding;
1196
1197 out:
1198         splice_shrink_spd(&spd);
1199         return ret;
1200 }
1201
1202 static ssize_t relay_file_splice_read(struct file *in,
1203                                       loff_t *ppos,
1204                                       struct pipe_inode_info *pipe,
1205                                       size_t len,
1206                                       unsigned int flags)
1207 {
1208         ssize_t spliced;
1209         int ret;
1210         int nonpad_ret = 0;
1211
1212         ret = 0;
1213         spliced = 0;
1214
1215         while (len && !spliced) {
1216                 ret = subbuf_splice_actor(in, ppos, pipe, len, flags, &nonpad_ret);
1217                 if (ret < 0)
1218                         break;
1219                 else if (!ret) {
1220                         if (flags & SPLICE_F_NONBLOCK)
1221                                 ret = -EAGAIN;
1222                         break;
1223                 }
1224
1225                 *ppos += ret;
1226                 if (ret > len)
1227                         len = 0;
1228                 else
1229                         len -= ret;
1230                 spliced += nonpad_ret;
1231                 nonpad_ret = 0;
1232         }
1233
1234         if (spliced)
1235                 return spliced;
1236
1237         return ret;
1238 }
1239
1240 const struct file_operations relay_file_operations = {
1241         .open           = relay_file_open,
1242         .poll           = relay_file_poll,
1243         .mmap           = relay_file_mmap,
1244         .read           = relay_file_read,
1245         .llseek         = no_llseek,
1246         .release        = relay_file_release,
1247         .splice_read    = relay_file_splice_read,
1248 };
1249 EXPORT_SYMBOL_GPL(relay_file_operations);