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